Autonomous AI Security Auditing of Webmin: 507 Newly Discovered Vulnerabilities An experienced security researcher ran a fully autonomous Codex campaign against Webmin 2.670 that produced 507 canonical, reproduced vulnerability findings across 44,297 tracked files and 476,216 Perl/CGI lines, with 524 applicability paths carrying CVSS 4.0 Base assessments. The campaign consumed 8,347,303,160 model tokens, which at the published 19 November 2026 Standard API prices for gpt-5.6-sol would cost US$4,577.83, and the author judged it captured over 90% of classic defects at the pinned revision, possibly every XSS. The catalogue also records 567 confirmed historical issues and 95 hardening observations, and the author reported a sharp rise in AI-produced findings within the past two months. Abstract An experienced security researcher translated techniques developed before modern AI into a fully autonomous Codex campaign against Webmin. The campaign delivered 507 canonical, reproduced findings at a pinned Webmin 2.670 revision with 44,297 tracked files, 476,216 Perl/CGI lines overall, and 386,178 production code lines measured by cloc. Its 524 applicability paths have CVSS 4.0 Base assessments and explicit conditions. Execution consumed 8,347,303,160 model tokens; the same token mix at the current Standard API rates would cost about US$4,578. Among paths without recorded special conditions, 48 of 73 XSS and 38 of 51 privilege-escalation paths match module gates present in at least 80% of 6,756 sampled host profiles. The catalogue also records 567 confirmed historical issues and 95 hardening observations. In the author’s judgment, the audit captured over 90% of classic defects at the pinned revision, possibly every XSS. The author has observed a sharp rise in AI-produced findings within the past two months, especially in recent weeks. This case study records what a fully autonomous, researcher-designed campaign can achieve at this rapidly changing point. Introduction Webmin exposes a broad administrative interface through core services and optional modules. Security impact depends on the route into a module, the permissions held by the acting principal, and deployment-specific services and options. This case study documents the output of a fully autonomous Codex campaign designed by an experienced researcher. Its version-bound catalogue records independently fixable findings, verified applicability paths, explicit prerequisites, CVSS vectors, source links, and a revision-bound inclusion rule. Later independent discoveries at the same revision can test the author’s assessment of near-saturation. Study design and evidence | Item | Audited scope | |---|---| | Audited target | Webmin 2.670 source at 3a2b2b8671c0 | | Units | 507 canonical findings; 526 reproduced catalogue records | | Host-profile sample | 6,756 usable module enumerations of 7,684 profile files | | Inclusion rule | A reproduced verification at the pinned revision | | Exclusions | Historical-only candidates and separate duplicate entries | Full audited source revision: 3a2b2b8671c0669c129f5275b3dd1853106ae02b . At that revision, the Git tree contains 44,297 tracked files 91.0 MB of blob content . Among them are 3,806 .pl , .pm , and .cgi files containing 476,216 lines, and 126 module.info manifests. Separately, cloc 2.04 counts 386,178 nonblank, noncomment code lines in 3,731 production files. That count includes Perl, JavaScript, CSS, shell, HTML, C, Python, and batch files, and excludes test, vendored Perl, demo, localization, documentation, and repository-metadata directories. AI-assisted research context The author adapted security-auditing techniques developed before modern AI into a role-based, multi-worker Codex system. After the objective was set, the system autonomously searched, analyzed, tested, reconciled, and recorded findings. The author was presented with the resulting catalogue. The campaign used the Daybreak Blue routing alias gpt-daybreak-blue-latest at xhigh reasoning effort 7,10 . The exact task design, orchestration, and lab configuration remain unpublished as the author’s long-developed research methods. The execution records contain 8,312,322,763 input tokens, including 8,158,630,784 cached tokens, and 34,980,397 output tokens: 8,347,303,160 tokens in total. At the published 19 November 2026 Standard API prices for gpt-5.6-sol , the current target of the Daybreak Blue alias, the recorded token mix would cost US$4,577.83 US$4.00 per million uncached input, US$0.40 cached input, and US$20.00 output 10,11 . This is an API-equivalent model-token estimate. We found no earlier publicly documented project-wide AI-assisted Webmin audit in its official security notices or GitHub advisories 8,9 . Finding identity and verification The target is a pinned source checkout. We selected 526 records with a reproduced verification at that revision and excluded later-only and historical-only observations. Identity reconciliation consolidated 19 duplicate records into 507 canonical findings. One finding represents an independently fixable defect; one applicability scenario represents a distinct route or configuration-dependent manifestation. A scenario enters the scored population only when it links to a selected verification. Each entry states the observed behavior and any recorded limits; raw verification records remain private. Of the 507 selected canonical records, 475 have an audited verification described as runtime, HTTP, or focused execution, while 32 use historical-observation review with latest-stable source and evidence reconciliation. These evidence classes are distinct; a reproduced status does not imply that every path was exercised in a live browser or installed package. The audit did not establish that every finding affects every official 2.670 package: a separate release source-divergence review found differences between the tag and published artifacts. That review is retained privately. Source revisions The audited source is 3a2b2b8671c0669c129f5275b3dd1853106ae02b . The checked public upstream source is 0dd66e3d5b5c1fa61212c38d7c8df0d283bb88b8 https://github.com/webmin/webmin/commit/0dd66e3d5b5c1fa61212c38d7c8df0d283bb88b8 . Evidence-backed public fix mappings appear in the appendix. CVE cross-references Published CVEs were known before the campaign and served as background, not as audit discoveries. The Known Bugs Webmin project currently links 109 distinct CVE identifiers: 107 published and two rejected. The full reference set appears in the appendix. None of the 507 audited-revision findings has a recorded CVE identifier. Classification and scoring The recorded weakness: , boundary: , and impact: tags determine the primary display section and path families. The primary section is a deterministic index; a finding may span several weaknesses or impacts. Reviewed applicability profiles record exposure modules, actor routes, verification links, and prerequisites. We score each reviewed path linked to a reproduced finding with the CVSS 4.0 Base metric set 1 , retaining the vector and metric rationales in the public data. Evidence references remain in the private scorecard. CVSS measures impact and exploit conditions for the documented path; the host-profile signal is kept separate. Host-profile sample and exposure gate The saved inventory was profiled on 2026-11-17 using anonymous version and public-asset module probes. Its inputs came from the project’s Censys, Netlas, and other known-host inventory streams. The unit is a profile file for a target endpoint; the corpus was not sampled at random and may contain several endpoints associated with one installation. No exploit requests were sent to these hosts as part of the exposure calculation. | Profile-selection step | Count | |---|---| | Saved target profiles | 7,684 | | Unavailable module enumerations | −13 | | Completed enumerations | 7,671 | | Only automatically included modules detected | −915 | | Usable module enumerations | 6,756 | Of the usable files, 2,892 have a detected 2.670 version signal, 962 have a different detected version, and 2,902 carried a legacy 2.670 fallback label when the version probe returned no evidence. Those fallback labels were excluded from detected-version counts and vulnerability inference. The accompanying webmin-fingerprint v0.1.0 emits null for an unobserved Webmin version. Results and interpretation The primary index places 140 findings under XSS and HTML injection, 116 under command and data injection, and 106 under access control and ACL bypass. These are display classes, not exclusive weakness counts or a comparison of deployment risk. The catalogue below keeps the component and path-specific conditions visible for each entry. All listed findings have a recorded reproduction against the audited revision. The imported severity labels are replaced by reviewed CVSS 4.0 Base assessments for 524 of 524 reviewed paths across 507 of 507 canonical findings. Each displayed score includes its vector and is this research project’s assessment, not a vendor or NVD rating. CVSS Base excludes the host-profile footprint. The published scorecard is tied to the audited revision. The per-entry percentage is the share of usable profiles where the exposure module’s unauthenticated asset probe was positive; core paths are assigned all usable profiles by definition of the exposure gate. Alternative path modules are combined as a union, counting each profile once. Only paths linked to a finding verification at the audited revision are counted; 11 unlinked alternative scenarios are omitted. Specific ACLs, options, external services, attacker control, and victim action are generally not observable in these profiles, so the displayed module signal is a screening estimate, not a measured vulnerability rate or a statistical upper bound for all Webmin installations. The sample is assembled from known hosts and is not representative of the global installed base; version applicability is not inferred from the anonymous version signal. Applicability profiles are present and reviewed for all 507 entries. The condition lines name prerequisites supported by the canonical finding evidence; the host corpus does not record whether a given installation meets them. N/A 13 entries marks installer, development, or unprobed-module paths that the running-host module corpus cannot measure. CVSS assessment coverage The 507 findings have 524 reviewed applicability paths. 524 paths have reviewed CVSS 4.0 Base assessments. Path ratings: 10 Critical, 253 High, 248 Medium, 10 Low, 3 None. CVSS Base scores reflect the documented exploit path in the affected configuration; they do not include the host-profile footprint. Vectors and metric rationales appear in public data; source reports remain private. Multiple paths on one finding can have different scores. A score of 0.0 marks a reproduced behavior without a demonstrated CVSS security impact, including the blocked reset-token redemption in WEBMIN-0916. 14 operator-workflow paths have no distinct attacker-route tag; their CVSS privilege metric models the authorized configuration action. 250 paths have no impact: taxonomy tag; their impact metrics were assessed from the finding narratives and linked evidence. CVSS is owned by FIRST and used by permission. See the FIRST CVSS v4.0 specification https://www.first.org/cvss/v4.0/specification-document for the scoring definitions. Individual paths relevant to one-click RCE The listed XSS browser-entry paths and privilege-escalation actions were proven end to end individually at the audited Webmin 2.670 revision. A compatible pair on the same host and session can form a one-click RCE route. The tables include reviewed paths with no recorded additional prerequisites. Rank is a dense rank of exposure-gate counts; CVSS orders ties for display. Find each full entry below by its ID. The exposure gate covers at least 80% of usable profiles for 48/73 XSS paths 48 findings and 38/51 privilege-escalation paths 38 findings . Module gates use positive asset probes; the one core path in each family is assigned to all usable profiles. If all completed enumerations are kept in the denominator 7,671 , the same 80% threshold is met by 24 XSS and 18 privilege-escalation paths. XSS paths without recorded special conditions 73 | Rank | Finding ID | Exposure gate | CVSS-B | Gate share | |---|---|---|---|---| | 1 | WEBMIN-0518 | Core | 5.3 Medium | 100.00% 6,756/6,756 | | 2 | WEBMIN-0048 | MySQL | 5.3 Medium | 97.10% 6,560/6,756 | | 2 | WEBMIN-0628 | MySQL | 5.1 Medium | 97.10% 6,560/6,756 | | 3 | WEBMIN-2443 | Net | 4.8 Medium | 95.80% 6,472/6,756 | | 4 | WEBMIN-0090 | Servers | 5.3 Medium | 95.66% 6,463/6,756 | | 4 | WEBMIN-0436 | Servers | 5.1 Medium | 95.66% 6,463/6,756 | | 4 | WEBMIN-0773 | Servers | 5.1 Medium | 95.66% 6,463/6,756 | | 5 | WEBMIN-0435 | Useradmin | 5.3 Medium | 95.63% 6,461/6,756 | | 5 | WEBMIN-0236 | Useradmin | 5.1 Medium | 95.63% 6,461/6,756 | | 6 | WEBMIN-0273 | Postfix | 5.3 Medium | 95.40% 6,445/6,756 | | 6 | WEBMIN-0590 | PHP Configuration | 5.1 Medium | 95.40% 6,445/6,756 | | 7 | WEBMIN-0598 | Package Updates | 5.1 Medium | 95.00% 6,418/6,756 | | 8 | WEBMIN-0557 | Filemin | 5.1 Medium | 94.83% 6,407/6,756 | | 8 | WEBMIN-0785 | Filemin | 4.8 Medium | 94.83% 6,407/6,756 | | 9 | WEBMIN-0430 | Bind8 | 5.1 Medium | 94.63% 6,393/6,756 | | 10 | WEBMIN-0917 | Dovecot | 5.1 Medium | 94.57% 6,389/6,756 | | 11 | WEBMIN-0034 | Mailboxes | 5.3 Medium | 94.32% 6,372/6,756 | | 11 | WEBMIN-0870 | Mailboxes | 5.3 Medium | 94.32% 6,372/6,756 | | 11 | WEBMIN-0336 | Mailboxes | 5.1 Medium | 94.32% 6,372/6,756 | | 12 | WEBMIN-0855 | Fail2Ban | 5.1 Medium | 93.75% 6,334/6,756 | | 13 | WEBMIN-0007 | Apache | 5.3 Medium | 92.07% 6,220/6,756 | | 13 | WEBMIN-0363 | Apache | 5.1 Medium | 92.07% 6,220/6,756 | | 13 | WEBMIN-0532 | Apache | 5.1 Medium | 92.07% 6,220/6,756 | | 13 | WEBMIN-0605 | Apache | 5.1 Medium | 92.07% 6,220/6,756 | | 14 | WEBMIN-0172 | Fetchmail | 5.1 Medium | 87.30% 5,898/6,756 | | 15 | WEBMIN-0218 | Firewall | 5.1 Medium | 86.68% 5,856/6,756 | | 15 | WEBMIN-0264 | Firewall | 5.1 Medium | 86.68% 5,856/6,756 | | 15 | WEBMIN-0542 | Firewall | 5.1 Medium | 86.68% 5,856/6,756 | | 16 | WEBMIN-0386 | Pam | 5.1 Medium | 86.04% 5,813/6,756 | | 17 | WEBMIN-0159 | Raid | 7.0 High | 85.98% 5,809/6,756 | | 18 | WEBMIN-0043 | Webalizer | 5.1 Medium | 85.60% 5,783/6,756 | | 19 | WEBMIN-0770 | Logviewer | 5.1 Medium | 85.24% 5,759/6,756 | | 20 | WEBMIN-0237 | LDAP Server | 5.1 Medium | 85.01% 5,743/6,756 | | 21 | WEBMIN-0361 | Iscsi Target | 5.1 Medium | 84.93% 5,738/6,756 | | 22 | WEBMIN-0464 | DHCP | 5.1 Medium | 84.87% 5,734/6,756 | | 22 | WEBMIN-0553 | DHCP | 5.1 Medium | 84.87% 5,734/6,756 | | 22 | WEBMIN-0635 | DHCP | 5.1 Medium | 84.87% 5,734/6,756 | | 22 | WEBMIN-0878 | DHCP | 5.1 Medium | 84.87% 5,734/6,756 | | 23 | WEBMIN-0309 | Samba | 5.1 Medium | 84.81% 5,730/6,756 | | 24 | WEBMIN-0548 | Shorewall6 | 5.1 Medium | 84.72% 5,724/6,756 | | 25 | WEBMIN-0073 | Shorewall | 5.1 Medium | 84.68% 5,721/6,756 | | 25 | WEBMIN-0132 | LDAP Client | 5.1 Medium | 84.68% 5,721/6,756 | | 25 | WEBMIN-0240 | Shorewall | 5.1 Medium | 84.68% 5,721/6,756 | | 25 | WEBMIN-0344 | Shorewall | 5.1 Medium | 84.68% 5,721/6,756 | | 25 | WEBMIN-0524 | Shorewall | 5.1 Medium | 84.68% 5,721/6,756 | | 26 | WEBMIN-0476 | Iscsi Tgtd | 6.8 Medium | 83.29% 5,627/6,756 | | 27 | WEBMIN-0452 | Bacula Backup | 5.3 Medium | 81.88% 5,532/6,756 | | 27 | WEBMIN-0721 | Bacula Backup | 5.1 Medium | 81.88% 5,532/6,756 | | 28 | WEBMIN-0359 | Proc | 2.4 Low | 20.43% 1,380/6,756 | | 29 | WEBMIN-0118 | Updown | 5.3 Medium | 5.60% 378/6,756 | | 29 | WEBMIN-0230 | Updown | 5.1 Medium | 5.60% 378/6,756 | | 30 | WEBMIN-0751 | Custom | 5.1 Medium | 5.52% 373/6,756 | | 31 | WEBMIN-0149 | Init | 5.1 Medium | 5.45% 368/6,756 | | 32 | WEBMIN-0293 | Webminlog | 5.1 Medium | 5.43% 367/6,756 | | 32 | WEBMIN-0427 | Webminlog | 5.1 Medium | 5.43% 367/6,756 | | 32 | WEBMIN-0734 | Webminlog | 5.1 Medium | 5.43% 367/6,756 | | 32 | WEBMIN-0954 | Webminlog | 5.1 Medium | 5.43% 367/6,756 | | 33 | WEBMIN-0031 | Webmin | 5.3 Medium | 5.39% 364/6,756 | | 34 | WEBMIN-0566 | Backup Config | 5.3 Medium | 5.31% 359/6,756 | | 34 | WEBMIN-0424 | Backup Config | 5.1 Medium | 5.31% 359/6,756 | | 35 | WEBMIN-0275 | Cpan | 5.1 Medium | 4.25% 287/6,756 | | 36 | WEBMIN-0015 | LDAP Users and Groups | 5.3 Medium | 4.17% 282/6,756 | | 37 | WEBMIN-0098 | Lpadmin | 5.3 Medium | 4.14% 280/6,756 | | 37 | WEBMIN-2405 | Lpadmin | 4.8 Medium | 4.14% 280/6,756 | | 38 | WEBMIN-0216 | Lvm | 5.1 Medium | 1.12% 76/6,756 | | 38 | WEBMIN-0298 | Lvm | 5.1 Medium | 1.12% 76/6,756 | | 38 | WEBMIN-0617 | Software | 5.1 Medium | 1.12% 76/6,756 | | 38 | WEBMIN-0651 | Software | 4.8 Medium | 1.12% 76/6,756 | | 39 | WEBMIN-0178 | Zones | 5.1 Medium | 0.64% 43/6,756 | | 40 | WEBMIN-0367 | Minecraft | 5.1 Medium | 0.04% 3/6,756 | | 40 | WEBMIN-0740 | Minecraft | 5.1 Medium | 0.04% 3/6,756 | | 41 | WEBMIN-0177 | Htpasswd File | 5.1 Medium | 0.01% 1/6,756 | | 42 | WEBMIN-0618 | Pptp Client | 5.1 Medium | 0.00% 0/6,756 | Privilege escalation paths without recorded special conditions 51 | Rank | Finding ID | Exposure gate | CVSS-B | Gate share | |---|---|---|---|---| | 1 | WEBMIN-0714 | MySQL | 8.7 High | 97.10% 6,560/6,756 | | 1 | WEBMIN-0880 | MySQL | 8.7 High | 97.10% 6,560/6,756 | | 1 | WEBMIN-0668 | MySQL | 7.1 High | 97.10% 6,560/6,756 | | 2 | WEBMIN-0029 | Spam | 7.1 High | 96.64% 6,529/6,756 | | 3 | WEBMIN-0437 | Servers | 7.1 High | 95.66% 6,463/6,756 | | 4 | WEBMIN-0398 | Postfix | 8.7 High | 95.40% 6,445/6,756 | | 4 | WEBMIN-0456 | PHP Configuration | 8.7 High | 95.40% 6,445/6,756 | | 4 | WEBMIN-0515 | Postfix | 8.7 High | 95.40% 6,445/6,756 | | 4 | WEBMIN-0686 | Postfix | 8.7 High | 95.40% 6,445/6,756 | | 4 | WEBMIN-0801 | Postfix | 7.1 High | 95.40% 6,445/6,756 | | 5 | WEBMIN-0313 | Mount | 8.7 High | 95.17% 6,430/6,756 | | 5 | WEBMIN-0536 | Mount | 7.1 High | 95.17% 6,430/6,756 | | 6 | WEBMIN-0967 | Usermin | 8.7 High | 94.75% 6,401/6,756 | | 7 | WEBMIN-0595 | Bind8 | 7.1 High | 94.63% 6,393/6,756 | | 8 | WEBMIN-0153 | Mailboxes | 7.1 High | 94.32% 6,372/6,756 | | 8 | WEBMIN-0401 | Mailboxes | 5.8 Medium | 94.32% 6,372/6,756 | | 9 | WEBMIN-0591 | Firewalld | 8.7 High | 93.04% 6,286/6,756 | | 9 | WEBMIN-0907 | Firewalld | 8.7 High | 93.04% 6,286/6,756 | | 10 | WEBMIN-0045 | Raid | 8.7 High | 85.98% 5,809/6,756 | | 10 | WEBMIN-0158 | Raid | 8.7 High | 85.98% 5,809/6,756 | | 10 | WEBMIN-0455 | Raid | 8.7 High | 85.98% 5,809/6,756 | | 10 | WEBMIN-0473 | Raid | 8.7 High | 85.98% 5,809/6,756 | | 10 | WEBMIN-0484 | Raid | 8.7 High | 85.98% 5,809/6,756 | | 10 | WEBMIN-0159 | Raid | 7.1 High | 85.98% 5,809/6,756 | | 11 | WEBMIN-0482 | Squid | 8.7 High | 85.82% 5,798/6,756 | | 11 | WEBMIN-0645 | Squid | 7.1 High | 85.82% 5,798/6,756 | | 12 | WEBMIN-0832 | Logviewer | 8.7 High | 85.24% 5,759/6,756 | | 12 | WEBMIN-2453 | Logviewer | 7.1 High | 85.24% 5,759/6,756 | | 13 | WEBMIN-0500 | Sarg | 8.7 High | 85.20% 5,756/6,756 | | 14 | WEBMIN-0351 | Openslp | 8.7 High | 85.08% 5,748/6,756 | | 15 | WEBMIN-0671 | Smf | 8.7 High | 84.92% 5,737/6,756 | | 16 | WEBMIN-0128 | DHCP | 7.1 High | 84.87% 5,734/6,756 | | 16 | WEBMIN-0227 | DHCP | 7.1 High | 84.87% 5,734/6,756 | | 16 | WEBMIN-0311 | DHCP | 7.1 High | 84.87% 5,734/6,756 | | 16 | WEBMIN-0529 | DHCP | 7.1 High | 84.87% 5,734/6,756 | | 17 | WEBMIN-0814 | Samba | 8.7 High | 84.81% 5,730/6,756 | | 18 | WEBMIN-0343 | Shorewall6 | 8.7 High | 84.72% 5,724/6,756 | | 19 | WEBMIN-0072 | Shorewall | 8.7 High | 84.68% 5,721/6,756 | | 20 | WEBMIN-0117 | Init | 7.1 High | 5.45% 368/6,756 | | 21 | WEBMIN-0256 | Webmin | 7.1 High | 5.39% 364/6,756 | | 22 | WEBMIN-0535 | Time | 8.7 High | 5.37% 363/6,756 | | 23 | WEBMIN-0680 | Proftpd | 8.7 High | 5.28% 357/6,756 | | 23 | WEBMIN-0027 | Proftpd | 7.1 High | 5.28% 357/6,756 | | 24 | WEBMIN-0499 | Cpan | 7.7 High | 4.25% 287/6,756 | | 25 | WEBMIN-0052 | Smart Status | 8.7 High | 4.23% 286/6,756 | | 26 | WEBMIN-0242 | Heartbeat | 8.7 High | 4.10% 277/6,756 | | 27 | WEBMIN-0379 | Lvm | 8.7 High | 1.12% 76/6,756 | | 28 | WEBMIN-0481 | Minecraft | 8.7 High | 0.04% 3/6,756 | | 28 | WEBMIN-0555 | Minecraft | 8.7 High | 0.04% 3/6,756 | | 28 | WEBMIN-0596 | Minecraft | 8.7 High | 0.04% 3/6,756 | | 28 | WEBMIN-0520 | Minecraft | 7.1 High | 0.04% 3/6,756 | Discussion A rapid change in security auditing In the author’s personal observation, the rate at which AI systems produce security findings has risen massively within the past two months, with the steepest visible increase in the past few weeks. The author cannot identify a single date for the change. The present case study captures a tipping point: an experienced researcher could encode earlier audit techniques into a fully autonomous Codex campaign and receive a large, source-linked, reproduced catalogue from a mature administrative application. The same methods and model capabilities are changing quickly, so this result serves as a dated record of what was possible at the time of the campaign. How complete was this audit? Drawing on years of security-research experience and the breadth of the campaign, the author believes it found more than 90% of the classic security defects in the pinned Webmin source, particularly XSS and authorization or privilege escalation flaws. The author considers it plausible that the campaign found every XSS present there. The project’s linked CVE reference set contains 107 published identifiers accumulated across Webmin releases; this campaign documented 507 independently fixable findings at the single audited revision. In the author’s judgment, the catalogue is close to complete for these classic classes. Some further issues may emerge, but the number of new findings against this exact revision should decline dramatically. Later independent reports can test that assessment by comparing each root cause and manifestation with the catalogue. The author plans to push a pull request to Webmin with fixes for all catalogue findings that still require changes on upstream master. Research ethics and automated assistance The host-profile analysis used previously saved anonymous asset and version probes; the paper reports aggregate counts and omits target addresses and raw profiles. The autonomous Codex campaign performed discovery, triage, validation, evidence organization, and classification. Inclusion is tied to recorded verification at the pinned revision. The author designed the campaign and assumes responsibility for the published claims. Related work The OWASP Web Security Testing Guide 2 organizes web-application tests around explicit controls and evidence of failure. Mu et al. 3 show why reproducibility details matter when interpreting reported vulnerabilities. Li et al. 5 find that project-scale LLM vulnerability detectors and agents can uncover distinct flaws while still suffering high false-discovery rates, incomplete context, and high costs. Sheng et al. 6 demonstrate automated discovery and reproduction. This case study keeps finding identity, applicability paths, and source observations separate; CWE 4 supplies a broader weakness vocabulary, and FIRST’s CVSS v4.0 1 supplies scoring definitions. Data and evidence availability This paper contains the included finding list, selection rule, reviewed path conditions, and links to the pinned public Webmin source where a direct source location was recorded. The accompanying public data archive public-data.zip contains structured data for 507 audited-revision findings, 524 included paths, 567 confirmed historical issues, and 95 hardening observations: prerequisites, exposure gates, CVSS vectors and rationales, ranking rows, repository-size measurements, anonymous module aggregates, campaign token totals, and mapped public fix commits. The companion MIT-licensed webmin-fingerprint release 10.5281/zenodo.24367586 https://doi.org/10.5281/zenodo.24367586 provides its source archive, an npx-ready npm package, and a Linux x86-64 binary, with SHA-256 checksums. Conclusion A researcher-designed autonomous AI campaign documented 507 canonical findings and 524 reviewed applicability paths at one pinned Webmin 2.670 source revision using 8,347,303,160 model tokens. It demonstrates the present capacity of a fully autonomous Codex audit designed from an experienced researcher’s methods. The author believes the catalogue approaches saturation for classic issues at this revision. Later independent findings will test that judgment. The complete cohort follows for finding-level review. Findings by primary type | Type | Canonical findings | |---|---| | Access control and ACL bypass | 106 | | Cross-site scripting and HTML injection | 140 | | Command and data injection | 116 | | File and path handling | 44 | | Information exposure and secrets | 41 | | Authentication and session handling | 14 | | Transport and cryptography | 19 | | Request forgery and origin checks | 4 | | Availability and race conditions | 7 | | Other security defects | 16 | | Total | 507 | The primary type is selected deterministically from the existing weakness: , boundary: , and impact: catalogue tags. A finding may have additional weaknesses or impacts; the section heading is an index, not a complete taxonomy. The final section collects uncommon or unclassified cases. Access control and ACL bypass 106 Core · ACL 8 WEBMIN-0143 — Global ACL save paths do not enforce ACL-edit permission CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . The ACL module’s user and group editors show global ACL controls only when the acting operator has ACL-edit permission, but save user.cgi and save group.cgi accept the posted acl security form marker and persist global ACL settings without rechecking that permission. These settings govern cross-module restrictions including filesystem root, dot-file visibility, read-only mode, and download-address policy. A direct request therefore crosses the delegation boundary even though the ordinary UI hides the controls. Profile-based footprint: 93.83% observed module signal 6,339/6,756 usable profiles; observed gate: acl . Recorded conditions: ACL delegate may save users or groups but is denied global ACL editing. nondefault Audited source: acl/edit user.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/acl/edit user.cgi L467 , acl/save user.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/acl/save user.cgi L371 . WEBMIN-0144 — Group save endpoint bypasses the groups permission CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . edit group.cgi rejects an operator whose ACL module permissions set groups=0, but save group.cgi has no equivalent server-side gate before it updates or creates groups and propagates their module lists. A crafted direct POST can therefore mutate group privileges despite the explicit denial, including adding a privileged module to all members of a chosen group. Profile-based footprint: 93.83% observed module signal 6,339/6,756 usable profiles; observed gate: acl . Recorded conditions: ACL delegate is denied group administration but retains access to the direct group-save endpoint. nondefault Audited source: acl/edit group.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/acl/edit group.cgi L12 , acl/save group.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/acl/save group.cgi L11 . WEBMIN-0145 — Client-certificate request handlers ignore the cert permission CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . The ACL module defines cert as the permission controlling whether a Webmin user can request a client certificate, but cert form.cgi and cert issue.cgi do not test it. An authenticated operator with cert=0 can still open the request form and reach issuance-handler validation. When Webmin’s client-certificate authority is configured and the request contains a valid browser-generated key, this exposes an alternative-login credential workflow that delegation policy explicitly denied. Profile-based footprint: 93.83% observed module signal 6,339/6,756 usable profiles; observed gate: acl . Recorded conditions: ACL delegate is denied certificate administration while a CA is configured. nondefault Audited source: acl/defaultacl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/acl/defaultacl L7 , acl/acl security.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/acl/acl security.pl L82 . WEBMIN-0224 — Password-policy delegates can administer the external identity store CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . The ACL module displays external SQL/LDAP identity-store controls only to principals with the sql permission, but its configuration-save, SQL-table creation, and LDAP-base provisioning handlers authorize on the separate pass permission. A constrained authenticated administrator with pass=1 and sql=0 can therefore cross the intended delegation boundary and modify the server-wide authentication backend. The latest-stable reproduction confirmed that this principal receives a normal redirect after posting to save sql.cgi. Profile-based footprint: 93.83% observed module signal 6,339/6,756 usable profiles; observed gate: acl . Recorded conditions: ACL delegate has password-policy permission but is denied external identity-store administration. nondefault Audited source: acl/edit sql.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/acl/edit sql.cgi L10 , acl/save sql.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/acl/save sql.cgi L10 . WEBMIN-0396 — Group ACL saving bypasses ACL/module restrictions and permits path traversal CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . save acl.cgi treats a non-empty acl group as sufficient to enter the group branch and checks only access.groups. It does not require access.acl, does not validate acl mod against the caller’s permitted modules, does not verify that the submitted group exists before the write, and constructs the ACL path from both request values. The local helper likewise writes or unlinks config directory/module/group.gacl without containment checks. This permits unauthorized group ACL changes and path traversal to an attacker-selected .gacl target. Profile-based footprint: 93.83% observed module signal 6,339/6,756 usable profiles; observed gate: acl . Recorded conditions: The escalation depends on configured ACL/module restrictions. nondefault Audited source: acl/save acl.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/acl/save acl.cgi , web-lib-funcs.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/web-lib-funcs.pl . WEBMIN-0522 — Group deletion can orphan anonymous access onto permissive module defaults CVSS 4.0 Base CVSS-B : 5.3 Medium · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:L/VA:N/SC:N/SI:N/SA:N . The direct deletion endpoints reject a Webmin user referenced by miniserv’s anonymous mapping, but delete group.cgi and delete groups.cgi require only group-management permission and pass each member to delete user without the same check. delete user removes the account, module assignment, and per-user ACL files while leaving the anonymous mapping untouched. A request admitted by the surviving mapping still receives the deleted username, and module ACL loading falls back to module defaults when its per-user ACL is absent. In the isolated latest-stable lab, a synthetic operator with groups=1, delete=0, and users=operator deleted a group containing the anonymous backing identity; the identity and group disappeared, the mapping remained, and an unauthenticated request to the mapped ACL index returned HTTP 200 with relevant defaults create=1, delete=1, groups=1, and users= . The path requires the stated anonymous-mapping configuration and delegated group-management permission. Profile-based footprint: 93.83% observed module signal 6,339/6,756 usable profiles; observed gate: acl . Recorded conditions: Anonymous Webmin access maps to a user who belongs to a group a delegate can delete. nondefault Audited source: acl/delete user.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/acl/delete user.cgi L15 , acl/delete group.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/acl/delete group.cgi . WEBMIN-0916 — Delegated ACL administrator can mint a reset token for an out-of-scope Unix account CVSS 4.0 Base CVSS-B : 0.0 None · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:N/VA:N/SC:N/SI:N/SA:N . The administrator-side reset handler authorizes only the Webmin account supplied as user or user acc, while a separate request value selects the Unix account placed in the token as uuser. Existence and sudo-capability checks establish that the Unix identity is valid, but do not establish that the delegated caller is allowed to reset it. The token consumer uses uuser in its Unix password-update branch. Consequently, an authenticated ACL administrator permitted to edit a low-privilege Webmin account can directly request a token for a different sudo-capable Unix account, crossing the configured target boundary. Frozen stable source retains this separation. The latest-stable lab confirmed issuance of a token with uuser=root for a caller authorized only over target, although the final password change was not reproduced because an independent hostname-mismatch check rejected that admin-generated token. Profile-based footprint: 100.00% assigned core gate 6,756/6,756 usable profiles . Core is a classification in the applicability model, not an observed module probe. Recorded conditions: A delegated ACL administrator may edit a low-privilege Webmin account but not the selected Unix account. nondefault ; The selected Unix account exists and satisfies the sudo-capability check; the demonstrated target was root. environmental Audited source: acl/forgot send.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/acl/forgot send.cgi L13 , forgot.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/forgot.cgi . WEBMIN-0988 — Safe users can inherit newly added permissive ACL capabilities CVSS 4.0 Base CVSS-B : 5.3 Medium · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:L/VA:N/SC:N/SI:N/SA:N . New safe users do not persist the negative global ACL marker intended to prevent future permission inheritance. The edit page sets negative=1 only in a local object, suppresses the global ACL form for safe users, and the safe save branch instead persists the root safe profile, which contains negative=0. Consequently, when an accessible module later adds an allow-by-default ACL key with a restrictive negative ACL value, the safe account inherits the permissive default. The frozen latest-stable container reproduction demonstrates this behavior and the negative=1 deny control. Profile-based footprint: 93.83% observed module signal 6,339/6,756 usable profiles; observed gate: acl . Recorded conditions: A safe user is granted a module that later gains a permissive allow-by-default ACL key. nondefault Audited source: acl/edit user.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/acl/edit user.cgi , acl/save user.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/acl/save user.cgi L394 . Core · Integration 3 WEBMIN-0333 — Configured HTTP tunnel can connect outside its allowed target and forward credentials CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:H/VI:N/VA:N/SC:N/SI:N/SA:N . The HTTP tunnel parses the request-supplied URL into connection host, port, and path separately from its configured target validation. A target that passes or bypasses the string check can direct the server to another authority. Fixed or variable upstream Basic credentials are then attached to that request, and upstream HTML is relayed through the Webmin endpoint. This creates authenticated SSRF with credential disclosure and same-origin active-content exposure. Profile-based footprint: 4.13% observed module signal 279/6,756 usable profiles; observed gate: tunnel . Recorded conditions: Tunnel is configured with a target restriction and upstream Basic credentials. nondefault ; An authenticated Tunnel user chooses another host reachable by the Webmin server. environmental Audited source: tunnel/link.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/tunnel/link.cgi , tunnel/tunnel-lib.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/tunnel/tunnel-lib.pl L12 . WEBMIN-0668 — Restricted MySQL module users can force-reset an arbitrary database account CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . The MySQL login endpoint accepts a per-user database ACL as sufficient authorization, but it takes login, password, and force from the request and calls the root-executed force-reset routine before updating that ACL. A restricted module user can therefore request a different database account such as root; the sink stops the database, starts it with grant checks disabled, and applies a password-change query. This crosses the intended per-user and per-database authorization boundary. Profile-based footprint: 97.10% observed module signal 6,560/6,756 usable profiles; observed gate: mysql . Recorded conditions: No additional deployment prerequisite is stored; actor and trigger details are in the finding narrative. WEBMIN-0722 — MySQL custom backup parameters bypass the authorized destination and backup user CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . The MySQL backup form accepts arbitrary mysqldump options. Although tokens are shell-quoted, options such as –result-file retain their program semantics and make the root-context mysqldump process write outside the selected backup path; the configured backup user applies only to the pipeline writer. Profile-based footprint: 97.10% observed module signal 6,560/6,756 usable profiles; observed gate: mysql . Recorded conditions: MySQL backup delegate has a destination restriction or demoted writer but can add custom mysqldump options. nondefault Audited source: mysql/backup form.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/mysql/backup form.cgi L154 , mysql/mysql-lib.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/mysql/mysql-lib.pl . Core · Webmin 4 WEBMIN-0032 — Mandatory update signatures fail open when GnuPG setup fails CVSS 4.0 Base CVSS-B : 7.7 High · CVSS:4.0/AV:N/AC:L/AT:P/PR:N/UI:P/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N . Both update-list and module-file verification are entered only when gnupg setup reports success. When setup fails, even documented mode 2 must check bypasses signature download and verification, then returns parsed metadata or a successful module check. An attacker able to influence the update channel can therefore defeat the requested authenticity policy. Profile-based footprint: 5.39% observed module signal 364/6,756 usable profiles; observed gate: webmin . Recorded conditions: Mandatory update-signature mode 2 is configured. nondefault ; GnuPG setup fails when an update is installed. environmental Audited source: webmin/webmin-lib.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/webmin/webmin-lib.pl L1200 . WEBMIN-0366 — IP-only proxy trust is not restricted to configured proxy sources CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . Miniserv restricts forwarded client-address headers to trusted proxies only when that list is non-empty. The access form requires a proxy entry for combined IP-and-certificate trust but permits IP-only trust with an empty list. In that configuration, any directly connected client can set the address subsequently used by per-user IP checks and logging. The Docker reproduction showed a loopback request denied by a per-user allow rule becoming accepted when it supplied the allowed address in X-Forwarded-For. Profile-based footprint: 100.00% assigned core gate 6,756/6,756 usable profiles . Core is a classification in the applicability model, not an observed module probe. Recorded conditions: IP-only proxy trust is enabled without a trusted-proxy source list. nondefault ; A client can reach MiniServ directly and supply a forged forwarded-address header. environmental Audited source: miniserv-lib.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/miniserv-lib.pl L244 , webmin/change access.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/webmin/change access.cgi . WEBMIN-0744 — Granular Webmin Configuration ACL does not protect lock termination endpoint CVSS 4.0 Base CVSS-B : 5.3 Medium · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:N/VA:L/SC:N/SI:N/SA:N . The Webmin Configuration index applies per-page allow/disallow ACLs only to visible links. kill lock.cgi does not enforce the lock-page ACL before enumerating active locks and signaling a matching PID. An authenticated module user restricted to another page can therefore directly terminate a root Webmin process that holds a valid lock, causing denial of service and risking interruption of a privileged configuration operation. Profile-based footprint: 5.39% observed module signal 364/6,756 usable profiles; observed gate: webmin . Recorded conditions: Webmin Configuration delegate is denied lock-page access but can reach kill lock.cgi. nondefault Audited source: webmin/index.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/webmin/index.cgi , webmin/kill lock.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/webmin/kill lock.cgi L4 . WEBMIN-2399 — TLS cipher-list rejection is not detected CVSS 4.0 Base CVSS-B : 5.1 Medium · CVSS:4.0/AV:N/AC:L/AT:N/PR:H/UI:N/VC:N/VI:L/VA:N/SC:N/SI:N/SA:N . MiniServ wraps Net::SSLeay::set cipher list in eval and reports only a non-empty Perl exception. Net::SSLeay reports an invalid cipher expression by returning false without throwing, so an intended cipher restriction can fail without the configured error being logged or acted upon, leaving the active TLS policy different from the administrator’s expectation. Profile-based footprint: 100.00% assigned core gate 6,756/6,756 usable profiles . Core is a classification in the applicability model, not an observed module probe. Recorded conditions: MiniServ is configured with a TLS cipher-list expression that the SSL library rejects without throwing. nondefault Audited source: miniserv-lib.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/miniserv-lib.pl . Module · Apache 3 WEBMIN-0008 — Client-controlled prior target bypasses alias directory restriction CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . Alias save logic allows a disallowed target when an old to request field is present and itself outside the allowed alias root. Because old to is a hidden client field rather than server-derived state, a restricted editor can forge it and save a new Alias or ScriptAlias pointing outside the permitted directory. The frozen newest-stable Docker harness reproduced acceptance of /etc/shadow while the simulated allowed root was /srv/allowed. Profile-based footprint: 92.07% observed module signal 6,220/6,756 usable profiles; observed gate: apache . Recorded conditions: Apache delegate has an alias target directory restriction and permission to save aliases. nondefault Audited source: apache/mod alias.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/apache/mod alias.pl L72 . WEBMIN-0305 — Direct virtual-server save bypasses Virtualmin ownership restriction CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . Apache module code identifies a virtual host managed by Virtualmin and withholds its address/root/name editor, while the generic directive saver also filters Virtualmin-owned fields on the server. However, apache/save vserv.cgi—the direct target of the withheld form—accepts an authorized direct request without evaluating is virtualmin domain. It then changes or deletes the virtual-host structure and writes DocumentRoot and ServerName. An authenticated Apache-module user who is allowed to edit that virtual host and has vaddr can therefore cross the intended Virtualmin ownership boundary. The path requires authenticated, target-specific module privileges and can cause service redirection, configuration desynchronization, or outage. Profile-based footprint: 92.07% observed module signal 6,220/6,756 usable profiles; observed gate: apache . Recorded conditions: Apache delegate has virtual-host editing and vaddr permission for a Virtualmin-owned host. nondefault Audited source: apache/virt index.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/apache/virt index.cgi L92 , apache/save virt.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/apache/save virt.cgi L19 . WEBMIN-2467 — Create-only Apache user can add a global wildcard Listen port CVSS 4.0 Base CVSS-B : 5.3 Medium · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:L/VA:N/SC:N/SI:N/SA:N . The Apache virtual-host creation endpoint authorizes the request with the create ACL, but its wildcard custom-port path can append a global numeric Listen directive without checking the separate global-options ACL or the directive allow/deny policy. A delegated create-only principal can therefore persist an additional all-interface listener that may become active on a later Apache reload or restart. The caller explicitly selects listening, the port is restricted to digits, and configuration activation occurs on a later Apache reload or restart. Profile-based footprint: 92.07% observed module signal 6,220/6,756 usable profiles; observed gate: apache . Recorded conditions: Apache delegate may create virtual hosts but is denied global Listen configuration. nondefault ; The creation request asks for a wildcard custom port and Apache is later reloaded for active exposure. environmental Audited source: apache/create virt.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/apache/create virt.cgi , apache/acl security.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/apache/acl security.pl . Module · At 1 WEBMIN-0545 — Restricted Webmin user can run another user’s At job CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . The single-job endpoint resolves an arbitrary request-supplied job ID from the complete At spool and then handles run and delete operations differently. Deletion obtains the caller’s module ACL and requires can edit user for the job owner, but the run branch immediately writes the stored job to a temporary executable and calls proc::safe process exec with the job owner’s UID and GID. Other reviewed paths filter or reject jobs through can edit user. Consequently, an authenticated At-module user restricted to a different Unix identity can bypass that restriction by directly posting id and run=1, trigger another user’s stored command early or repeatedly, and receive its output. A frozen latest-stable reproduction used a harmless root-owned synthetic job: the mode=3 session was denied by edit job.cgi, yet delete job.cgi executed the same job as UID 0. Profile-based footprint: 4.35% observed module signal 294/6,756 usable profiles; observed gate: at . Recorded conditions: At delegate can access the module but is denied control of the selected job owner’s account. nondefault Audited source: at/delete job.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/at/delete job.cgi L13 , at/edit job.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/at/edit job.cgi L13 . Consolidated reproduction records: webmin-0176 Restricted At-module user can force execution of another user’s queued job . Module · Bind8 4 WEBMIN-0017 — Find Free IPs bypasses per-zone authorization CVSS 4.0 Base CVSS-B : 5.3 Medium · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:L/VI:N/VA:N/SC:N/SI:N/SA:N . The Find Free IPs CGI resolves an arbitrary requested zone and reads its records when access.findfree is enabled, but does not apply can edit zone to the resolved object. Per-zone, per-view, and directory restrictions are therefore not enforced on this read workflow. A restricted authenticated module user can query record-derived address occupancy and host information for another configured zone. Profile-based footprint: 94.63% observed module signal 6,393/6,756 usable profiles; observed gate: bind8 . Recorded conditions: BIND delegate has Find Free IPs permission but is denied the requested zone by per-zone, view, or directory scope. nondefault ; The target BIND zone is configured and contains records from which address occupancy can be inferred. environmental Audited source: bind8/find free.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/bind8/find free.cgi L14 , bind8/bind8-lib.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/bind8/bind8-lib.pl . WEBMIN-0019 — BIND lifecycle handlers collapse distinct apply scopes CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:N/VA:H/SC:N/SI:N/SA:N . The BIND ACL models apply=2 as zone-only and apply=3 as whole-configuration-only. start.cgi and stop.cgi nevertheless authorize both modes with a truthiness test and immediately invoke start bind or stop bind. Hiding lifecycle links does not protect direct requests, so a scoped module user can control DNS service availability outside the granted operation class. Profile-based footprint: 94.63% observed module signal 6,393/6,756 usable profiles; observed gate: bind8 . Recorded conditions: BIND delegate has apply=2 zone-only or apply=3 configuration-only , without whole-server lifecycle permission. nondefault ; The BIND service is configured so a start or stop request affects DNS availability. environmental Audited source: bind8/acl security.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/bind8/acl security.pl L83 , bind8/start.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/bind8/start.cgi L12 . WEBMIN-0191 — DNSSEC permission bypass on direct zone signing CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . The BIND module has a dedicated dnssec permission, and normal DNSSEC key-management routes enforce it. The direct sign zone.cgi action checks only zone edit access before calling the signing routine. In the frozen 2.670 lab, dnssec=0 blocked the control page while the direct action reached signing and changed the zone file before failing on the intentionally unavailable signer. Profile-based footprint: 94.63% observed module signal 6,393/6,756 usable profiles; observed gate: bind8 . Recorded conditions: BIND delegate can edit a zone but is denied the separate DNSSEC permission. nondefault Audited source: bind8/sign zone.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/bind8/sign zone.cgi L12 . WEBMIN-0429 — Zone conversion bypasses per-zone and read-only ACLs CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . convert master.cgi and convert slave.cgi accept a requested zone and mutate its type/directives after checking only the broad destination-type capability. Neither handler enforces can edit zone or the read-only ACL. Frozen-stable execution changed an excluded zone from master to slave while the caller’s module ACL allowed only allowed.test and xss.test and set ro=1. Profile-based footprint: 94.63% observed module signal 6,393/6,756 usable profiles; observed gate: bind8 . Recorded conditions: BIND delegate has destination-type permission but is denied the selected zone or is read-only. nondefault Module · Change User 1 WEBMIN-2458 — Change Language and Theme locale ACL deny is ineffective CVSS 4.0 Base CVSS-B : 5.3 Medium · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:L/VA:N/SC:N/SI:N/SA:N . The Change Language and Theme ACL form offers an administrator a locale yes/no control, and the module’s mutation handler consults that permission. However, the ACL save callback copies only lang, theme, and pass. Since defaultacl supplies locale=1, submitting locale=0 produces no stored override and effective ACL loading restores allow. The affected user can therefore continue changing only that user’s locale or legacy date format contrary to the administrator’s selected policy. Profile-based footprint: 5.34% observed module signal 361/6,756 usable profiles; observed gate: change-user . Recorded conditions: Change Language and Theme delegate is explicitly denied locale changes. nondefault Module · Cluster Useradmin 1 WEBMIN-0567 — Cluster-useradmin feature switches hide controls without enforcing authorization CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . Cluster Users and Groups exposes configuration switches labeled as allowing user/group creation, user/group search, refresh, and synchronization, but the implementation consults them only when rendering index-page controls. The corresponding CGI handlers do not enforce the switches. The related conf host links setting likewise changes rendered links without protecting edit host.cgi. Consequently, an authenticated user with module access can directly request disabled functionality. The frozen 2.670 lab reproduced direct access to the actual create forms, search handlers, synchronization form, and refresh handler after all seven options were set to zero; no mutating form was submitted during the defensive test. Profile-based footprint: 4.17% observed module signal 282/6,756 usable profiles; observed gate: cluster-useradmin . Recorded conditions: Cluster Users and Groups feature switch denies an operation while the delegate retains module access. nondefault Audited source: cluster-useradmin/index.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/cluster-useradmin/index.cgi . Module · Cron 1 WEBMIN-0065 — Hourly scheduling policy is bypassable by a crafted save request CVSS 4.0 Base CVSS-B : 5.3 Medium · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:L/VA:N/SC:N/SI:N/SA:N . The hourly-only module and per-user ACL settings constrain only the rendered minute controls. The server-side parse times input routine accepts all mins or arbitrary minute arrays without checking hourly only. An authenticated Cron-module user who may create or edit a permitted target user’s job can therefore schedule an authorized command every minute or at multiple minutes per hour, bypassing the administrator’s frequency policy and potentially causing resource exhaustion or more persistent execution. Profile-based footprint: 96.86% observed module signal 6,544/6,756 usable profiles; observed gate: cron . Recorded conditions: Cron delegate is subject to the hourly-only scheduling restriction. nondefault Audited source: cron/cron-lib.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/cron/cron-lib.pl . Module · Custom 1 WEBMIN-2477 — Safe-mode ACL authorizes every configured custom command CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . The Custom Commands safe-mode ACL disables editing but sets cmds to a wildcard. When a user’s module ACL carries the safe marker, Webmin loads this template, and can run command treats the wildcard as authorization for every configured command. Assigning the module therefore exposes current and future custom commands—including commands configured for privileged accounts—rather than a least-privilege subset. Administrator assignment of the module remains a required precondition. Profile-based footprint: 5.52% observed module signal 373/6,756 usable profiles; observed gate: custom . Recorded conditions: Custom Commands is assigned to a user through the safe-mode ACL profile. nondefault ; At least one configured command runs with privileges beyond that user. environmental Audited source: custom/safeacl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/custom/safeacl L1 , web-lib-funcs.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/web-lib-funcs.pl L5168 . Module · DHCP 2 WEBMIN-0129 — Alternate MAC encoding bypasses DHCP host uniqueness policy CVSS 4.0 Base CVSS-B : 5.3 Medium · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:L/VA:N/SC:N/SI:N/SA:N . save host.cgi checks a submitted hardware address for a duplicate by comparing its raw text with stored values, and only afterward normalizes dash-delimited or unformatted addresses into colon notation. Because multiple accepted spellings identify the same MAC address, a delegated host creator can submit an alternate spelling that passes uniq hst and is then persisted identically to an existing reservation. The newest-stable run with host-create permission and uniqueness enforcement enabled saved two 00:11:22:33:44:55 entries. The demonstrated impact is DHCP reservation integrity and potential availability loss, not privilege escalation or command execution. Profile-based footprint: 84.87% observed module signal 5,734/6,756 usable profiles; observed gate: dhcpd . Recorded conditions: DHCP host-MAC uniqueness enforcement is enabled. nondefault Audited source: dhcpd/save host.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/dhcpd/save host.cgi L65 . WEBMIN-0529 — Delegated DHCP array-option editing permits root-owned configuration directive injection CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . In save options.cgi, a custom option whose definition begins with array follows a branch that performs no validation or quoting of cval N. The value is handed to directive lines, which concatenates unquoted values verbatim. An authenticated DHCP-module user who is meant to edit only an allowed subnet, host, group, or shared network can include a semicolon and additional syntax so Webmin writes extra statements into the root-owned DHCP configuration. The frozen 2.670 runtime reproduced this with a subnet-only account lacking global and apply privileges: cval 0 set to 1; authoritative produced option lab-array 1; authoritative; in a UID-0-owned dhcpd.conf. This establishes persistent arbitrary DHCP directive injection and a delegated authorization-boundary bypass. The test did not execute a DHCP daemon command or establish direct OS-root command execution. Profile-based footprint: 84.87% observed module signal 5,734/6,756 usable profiles; observed gate: dhcpd . Recorded conditions: No additional deployment prerequisite is stored; actor and trigger details are in the finding narrative. Audited source: dhcpd/save options.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/dhcpd/save options.cgi L136 , dhcpd/dhcpd-lib.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/dhcpd/dhcpd-lib.pl L423 . Module · Fdisk 1 WEBMIN-0779 — fdisk mutation endpoint does not enforce partition safety and backend policy CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . The fdisk edit pages suppress mutation controls when a partition is mounted or when a GPT partition is represented as read-only because parted is unavailable or module mode is fdisk. The authoritative save part.cgi endpoint rechecks only the module user’s disk selection. It does not re-evaluate device status, require the editable marker value, or enforce the configured GPT backend before invoking delete partition or change type. An authenticated user who already has fdisk edit access to the selected disk can bypass these intended state/backend constraints with a crafted direct request, creating a credible integrity and availability risk. The caller must already have module edit permission for the selected disk. Profile-based footprint: 95.60% observed module signal 6,459/6,756 usable profiles; observed gate: fdisk . Recorded conditions: Fdisk editor targets a mounted partition or a GPT partition whose normal UI disables mutation. environmental ; The delegate still has fdisk edit permission for the selected disk. nondefault Audited source: fdisk/edit part.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/fdisk/edit part.cgi , fdisk/edit disk.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/fdisk/edit disk.cgi . Module · Filemin 3 WEBMIN-0526 — Direct file-edit requests bypass the per-user MIME edit ACL CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . Filemin exposes allowed for edit as a per-user ACL describing MIME types permitted for editing, but the value is used only to decide whether the directory listing shows an edit link. The edit and save endpoints enforce path containment and filesystem permissions but do not compare the target MIME type with allowed for edit. A user can directly invoke those endpoints to read or modify an excluded file inside an otherwise allowed directory. Frozen 2.670 source retains this split, and an isolated authenticated latest-stable run reproduced it: the UI omitted the edit action for blocked.bin under an application-json-only ACL, while a direct save file.cgi POST changed the binary file from BEFORE-B01009 to AFTER-B01009. Profile-based footprint: 94.83% observed module signal 6,407/6,756 usable profiles; observed gate: filemin . Recorded conditions: Filemin delegate has a MIME edit allowlist that excludes an otherwise accessible file. nondefault Audited source: filemin/acl security.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/filemin/acl security.pl , filemin/filemin-lib.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/filemin/filemin-lib.pl . WEBMIN-0876 — Direct editor endpoints bypass the MIME editing allowlist CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . The File Manager ACL labels allowed for edit as the MIME types permitted for the built-in editor. The listing code consults that value when deciding whether to render an edit action, but edit file.cgi and save file.cgi enforce only the allowed path and underlying file permissions. A user can request those handlers directly for a disallowed MIME type and read or replace the file. The controlled 2.670 reproduction configured text/plain as the sole allowed edit type, confirmed that an image/png target had no edit action, then read its marker and overwrote it through the direct handlers. Profile-based footprint: 94.83% observed module signal 6,407/6,756 usable profiles; observed gate: filemin . Recorded conditions: Filemin MIME edit allowlist excludes a file within the delegate’s allowed path. nondefault Audited source: filemin/acl security.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/filemin/acl security.pl , filemin/filemin-lib.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/filemin/filemin-lib.pl . WEBMIN-0886 — Strict ownership ACL permits writable files owned by other users CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:L/SC:N/SI:N/SA:N . The strict File Manager ACL is described as allowing deletion and movement only for files owned by the selected Unix user. Its gate uses Perl’s -w predicate instead of comparing file ownership, so group-writable or ACL-writable files owned by another UID pass. The frozen 2.670 container demonstrated UID 1001 passing the gate and renaming a file owned by UID 1002. Profile-based footprint: 94.83% observed module signal 6,407/6,756 usable profiles; observed gate: filemin . Recorded conditions: Filemin delegate uses strict ownership mode for a selected Unix account. nondefault ; A different-owned file is writable by that account through group or filesystem ACL permissions. environmental Audited source: filemin/filemin-lib.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/filemin/filemin-lib.pl L69 , filemin/lang/en https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/filemin/lang/en L135 . Module · Filter 1 WEBMIN-0619 — Usermin mail-filter condition injection enables same-account command execution CVSS 4.0 Base CVSS-B : 5.3 Medium · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:L/VI:L/VA:L/SC:N/SI:N/SA:N . The simple header-condition save path does not reject carriage returns or line feeds. When regex mode is disabled it only applies quotemeta if a fixed regular-expression metacharacter set is present; line breaks, backticks, and equals signs are outside that set. The downstream Procmail recipe serializer concatenates the condition directly and preserves embedded line breaks. An authenticated Usermin Filter module user can therefore inject a new Procmail action or variable assignment with command substitution, then trigger it with matching mail. Commands run as the corresponding Unix account because the module switches to the remote user before loading and writing filter data. This bypasses module-only restrictions but is not direct root execution. Profile-based footprint: 87.39% observed module signal 5,904/6,756 usable profiles; observed gate: filter . Recorded conditions: A Usermin Filter user can save a line-break-bearing condition and a matching message is delivered. environmental Audited source: filter/save.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/filter/save.cgi L86 , procmail/procmail-lib.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/procmail/procmail-lib.pl L231 . Module · Firewall 4 WEBMIN-0217 — Restricted firewall user can move rules with disallowed actions CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . The move-selected dispatcher performs table-level authorization but does not call can jump for the indexed rules it relocates. A restricted firewall user can submit a rule index that the UI withholds and move a rule carrying a disallowed target action to another chain. Saved-file mode silently changes the policy an administrator may later apply; direct mode writes the changed table through iptables-restore immediately. The latest-stable lab safely used an inert save file and confirmed that a user limited to jumps=DROP moved an ACCEPT rule into INPUT. Profile-based footprint: 86.68% observed module signal 5,856/6,756 usable profiles; observed gate: firewall . Recorded conditions: Firewall delegate may move rules but has a jump-target restriction excluding the selected rule. nondefault Audited source: firewall/save policy.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/firewall/save policy.cgi , firewall/firewall-lib.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/firewall/firewall-lib.pl L431 . WEBMIN-0573 — Filtered-chain protection is bypassable through direct mutation endpoints CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . The firewall index treats filter chain matches as externally managed: it hides matching chains and suppresses controls for rules that jump to them. The mutation handlers do not repeat that policy. A delegated firewall user who has table and chain-operation ACLs can call save policy.cgi or a rule handler directly with the hidden chain name and change or remove its rules. This can defeat an administrator’s separation between Webmin-managed rules and protections managed by fail2ban or another firewall tool. Profile-based footprint: 86.68% observed module signal 5,856/6,756 usable profiles; observed gate: firewall . Recorded conditions: Firewall filter chain protects a chain while the delegate retains table and chain mutation rights. nondefault Audited source: firewall/index.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/firewall/index.cgi , firewall/save policy.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/firewall/save policy.cgi L13 . WEBMIN-0574 — Caller-controlled IP family selection crosses firewall module ACL boundaries CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . Privileged firewall handlers derive the target family from the version request parameter but derive permissions from the CGI’s current module name. A request to the firewall module with version=inet6 therefore loads IPv6 operations and data while retaining the firewall module ACL. A user assigned only firewall, with firewall6 unavailable or more restrictive, can use allowed firewall actions against IPv6 rules. Profile-based footprint: 86.68% observed module signal 5,856/6,756 usable profiles; observed gate: firewall . Recorded conditions: Firewall delegate can edit IPv4 rules but lacks comparable IPv6 firewall access. nondefault Audited source: firewall/firewall-lib.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/firewall/firewall-lib.pl L35 , firewall/save policy.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/firewall/save policy.cgi L5 . WEBMIN-0850 — IPv6 firewall address validation fails open CVSS 4.0 Base CVSS-B : 5.3 Medium · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:L/VA:N/SC:N/SI:N/SA:N . The IPv6 firewall library’s check ipmask routine calculates whether each comma-separated item is acceptable, but its invalid-item branch returns 1 and its completion path also returns 1. save rule6.cgi uses that return value as the validation gate for source and destination fields, then stores the normalized value. An authenticated Firewall-module user with rule-edit permission can therefore submit malformed address text that is accepted into rule state. A malformed saved rule can prevent the intended IPv6 firewall configuration from applying under an already privileged module account. Profile-based footprint: 86.68% observed module signal 5,856/6,756 usable profiles; observed gate: firewall . Recorded conditions: Firewall delegate can save an IPv6 rule with an invalid source or destination address. environmental Audited source: firewall/firewall6-lib.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/firewall/firewall6-lib.pl L48 , firewall/save rule6.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/firewall/save rule6.cgi L150 . Module · Firewall6 4 WEBMIN-0540 — Jump-target ACL bypass through custom rule arguments CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . A delegated firewall editor with a restricted jumps ACL can submit no parsed jump and place a long-form –jump target in the raw custom-arguments field. can jump authorizes the empty parsed target, and rule serialization appends the unchecked custom argument after known options. This permits firewall actions outside the administrator-defined target allowlist. Profile-based footprint: 83.53% observed module signal 5,643/6,756 usable profiles; observed gate: firewall6 . Recorded conditions: The boundary requires a configured jump-target restriction. nondefault Audited source: firewall/save rule6.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/firewall/save rule6.cgi , firewall/firewall-lib.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/firewall/firewall-lib.pl . WEBMIN-0541 — Bulk rule delete and move bypass per-rule jump authorization CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . The bulk delete and move branches trust submitted rule indexes and do not call can jump for each affected object. Although the rule list hides checkboxes for disallowed targets, a crafted request can mutate those rules. This violates the module’s delegated jump-target restriction and can affect firewall integrity and availability. Profile-based footprint: 86.68% observed module signal 5,856/6,756 usable profiles; observed gate: firewall . Recorded conditions: Firewall delegate can bulk-delete or move rules but is denied their jump target. nondefault Audited source: firewall/save policy.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/firewall/save policy.cgi L142 . WEBMIN-0593 — Request parameter selects IPv6 operations under IPv4 module authorization CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . Shared firewall CGI scripts authorize the module named by the request URL during init config, then get ipvx version accepts a version field from the request to select IPv4 or IPv6 implementation code. A user allowed to access firewall but denied firewall6 can therefore request /firewall/index.cgi?version=inet6 and enter IPv6 behavior while get module acl still resolves firewall permissions. If the firewall ACL grants rule-management capabilities, that user can act on an IP-family policy outside the denied firewall6 module assignment and can weaken or replace IPv6 filtering. Profile-based footprint: 86.68% observed module signal 5,856/6,756 usable profiles; observed gate: firewall . Recorded conditions: Firewall delegate is allowed IPv4 management but denied firewall6 management. nondefault Audited source: firewall/firewall-lib.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/firewall/firewall-lib.pl , firewall/index.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/firewall/index.cgi L7 . WEBMIN-0603 — IPv6 firewall operations can use the IPv4 firewall module authorization context CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . Webmin authorizes access using the module name derived from the CGI path, and feature ACL loading defaults to that same active module. The shared firewall endpoints then accept a request-controlled version value; a value ending in 6 selects firewall6-lib.pl even when the request remains under /firewall/. Consequently, an authenticated user globally allowed firewall but denied firewall6 can invoke IPv6 firewall behavior while retaining the firewall module’s table and action permissions. This crosses the global module-access boundary and can permit unauthorized IPv6 rule, policy, apply, or related operations when the corresponding firewall feature ACL is enabled. Profile-based footprint: 86.68% observed module signal 5,856/6,756 usable profiles; observed gate: firewall . Recorded conditions: Firewall delegate is allowed IPv4 management but denied firewall6 management. nondefault Audited source: firewall/firewall-lib.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/firewall/firewall-lib.pl L35 , web-lib-funcs.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/web-lib-funcs.pl . Module · Fsdump 2 WEBMIN-0494 — Restore-only fsdump users can select an arbitrary program for remote execution CVSS 4.0 Base CVSS-B : 8.7 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N . The filesystem-backup restore endpoint authorizes access with the restore ACL, while the module separately exposes cmds and extra ACL controls. For tar restores, the rmt input is neither hidden nor rejected when those command-oriented privileges are disabled. Webmin validates it as a single non-whitespace value and escapes it for the local shell, then passes it to GNU tar as –rmt-command. GNU tar supplies that value to the remote-shell transport as the program to start on the remote host. Consequently, a restore-only Webmin user who can authenticate to a remote target can select an arbitrary existing remote executable, including a shell, under the authenticated remote account. The local shell escaping is effective against local metacharacter injection but does not enforce the intended module authorization boundary for remote program selection. Profile-based footprint: 5.31% observed module signal 359/6,756 usable profiles; observed gate: fsdump . Recorded conditions: FSdump delegate can restore tar archives but is denied command and extra-option controls. nondefault ; Remote tar transport is used and the request supplies a custom rmt program. environmental Audited source: fsdump/restore.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/fsdump/restore.cgi L8 , fsdump/linux-lib.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/fsdump/linux-lib.pl . WEBMIN-0851 — Restricted filesystem-backup editors can execute a custom remote-shell program as root CVSS 4.0 Base CVSS-B : 8.7 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N . The Filesystem Backup module exposes create/edit permission separately from its before/after-command and extra-argument permissions, but its remote-shell parser accepts a crafted custom selector and an arbitrary executable path without consulting either command-related ACL. The value is persisted as the backup’s remote-shell command and supplied to GNU tar for a remote backup. Enabled backup jobs are created for root, so a restricted fsdump editor who can place or reference an executable can cause that executable to run with root privileges when the backup runs. Profile-based footprint: 5.31% observed module signal 359/6,756 usable profiles; observed gate: fsdump . Recorded conditions: An enabled remote backup job uses GNU tar and a custom remote-shell executable. nondefault ; The restricted editor can place or reference the selected executable. environmental ; The scheduled or manual root-run backup job executes afterward. environmental Audited source: fsdump/save dump.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/fsdump/save dump.cgi , fsdump/fsdump-lib.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/fsdump/fsdump-lib.pl L326 . Module · Init 4 WEBMIN-0277 — Run-only Bootup ACL can write Upstart and systemd service definitions CVSS 4.0 Base CVSS-B : 8.7 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N . Bootup ACL value 2 is the module’s start-only role, but both Upstart and systemd save handlers authorize any nonzero value before allowing create, replace, and delete operations on root service definitions. A limited Webmin user can place attacker-chosen commands into a service definition and start it through the permissions they already hold, converting run-only module access into root-level command execution. Profile-based footprint: 5.45% observed module signal 368/6,756 usable profiles; observed gate: init . Recorded conditions: The finding is specific to a non-default run-only Bootup ACL. nondefault Audited source: init/acl security.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/init/acl security.pl L10 , init/save upstart.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/init/save upstart.cgi . WEBMIN-0346 — Run-only init-module users can create arbitrary root systemd services CVSS 4.0 Base CVSS-B : 8.7 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N . The run-only ACL tier is intended to control existing boot actions without edit rights, as shown by save action.cgi requiring bootup==1 for modifications. The systemd save endpoint instead accepts any nonzero bootup value and allows new services with request-controlled atstart content. create systemd service writes that content to ExecStart in a privileged unit. A bootup=2 user can therefore configure a service command that systemd runs as root, exceeding the run-only authorization boundary. Profile-based footprint: 5.45% observed module signal 368/6,756 usable profiles; observed gate: init . Recorded conditions: The finding is specific to a non-default run-only Bootup ACL. nondefault Audited source: init/acl security.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/init/acl security.pl L10 , init/save systemd.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/init/save systemd.cgi L6 . WEBMIN-0495 — Run-only init ACL permits privileged launchd creation and configuration CVSS 4.0 Base CVSS-B : 8.7 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N . The init module defines bootup ACL value 2 as run-only, but edit launchd.cgi and save launchd.cgi authorize with a non-zero truth test rather than requiring value 1. A run-only delegated user can therefore open the launchd creation/configuration form and reach creation, raw plist overwrite, reload, and deletion branches. On a macOS launchd host these operations manage root-owned system daemons; choosing the program for a new daemon or rewriting a plist can execute attacker-selected commands as root. Profile-based footprint: 5.45% observed module signal 368/6,756 usable profiles; observed gate: init . Recorded conditions: The finding is specific to a non-default run-only Bootup ACL. nondefault WEBMIN-0759 — Run-only init ACL can mutate root systemd units CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . The init ACL defines bootup value 2 as run only, while full mutation is value 1. In frozen stable, edit systemd.cgi and save systemd.cgi authorize with a nonzero test, so a run-only account can open the unit editor and submit creation or replacement data. The save path passes a supplied start command to create systemd service, which writes a root-owned unit ExecStart and reloads systemd. This crosses the documented ACL boundary and can convert delegated service-control access into persistent root command execution. Profile-based footprint: 5.45% observed module signal 368/6,756 usable profiles; observed gate: init . Recorded conditions: The finding is specific to a non-default run-only Bootup ACL. nondefault Audited source: init/mass systemd.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/init/mass systemd.cgi L58 . Module · Iscsi Client 1 WEBMIN-0356 — Prefix-only numeric validation permits Open-iSCSI configuration directive injection CVSS 4.0 Base CVSS-B : 7.1 High · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N . The save handler for login retry count, maximum queued commands, and device queue depth validates only a leading positive integer with /^ 1-9 \d /. It then sends the complete request value to save directive, which concatenates it into an iscsid.conf line. A crafted value containing a valid numeric prefix and newline therefore creates an additional Open-iSCSI directive, allowing a module user to bypass the form’s intended field restrictions and alter authentication, startup, or other client behavior. Profile-based footprint: 4.11% observed module signal 278/6,756 usable profiles; observed gate: iscsi-client . Recorded conditions: iSCSI Client delegate can save the affected numeric settings with a numeric prefix and newline. environmental Audited source: iscsi-client/save iscsi.cgi https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/iscsi-client/save iscsi.cgi L19 , iscsi-client/iscsi-client-lib.pl https://github.com/webmin/webmin/blob/3a2b2b8671c0669c129f5275b3dd1853106ae02b/iscsi-client/iscsi-client-lib.pl L70 . Module · LDAP Client 1 WEBMIN-0493 — NSS LDAP filter persistence fails open CVSS 4.0 Base CVSS-B : 5.3 Medium · CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:L/VA:N/SC:N/SI:N/SA:N . When saving a per-database filter for the legacy NSS LDAP client, save base.cgi appends the filter text to $file instead of $base. The script subsequently persists only $base as nss base