Gone in three hours: What ARTEX does with your credentials The FBI's November 13 advisory FLASH-20261113-001 reports that threat actors using ARTEX, an open-source multi-agent pentesting framework written in Go, can move from initial targeting to privileged access in under three hours, with malicious use documented since at least August 2026. The developer pulled the GitHub repository on October 8, but forks and a commercial version kept ARTEX circulating, and researchers counted more than 1,600 publicly exposed ARTEX consoles by early November. The FBI maps the activity to 43 ATT&CK techniques across at least ten market segments, including one campaign that exfiltrated personal information on about 67,000 people and a cloud backup supply chain compromise that exposed more than 2,250 downstream tenants. Threat actors can now go from initial targeting to privileged access in under three hours. Your change advisory board takes longer to approve a firewall rule. That figure comes from the FBI’s November 13 advisory, FBI FLASH-20261113-001 https://www.ic3.gov/CSA/2026/261113.pdf , about a framework called ARTEX. The read as a recovering CISO: all the techniques in the alert are familiar, but the clock has changed. Every credential you left lying around just got easier to pick up. ARTEX is an open-source, multi-agent pentesting framework written in Go. An operator types in an objective in plain language. LLM-backed agents turn it into commands, load the proper scripts, and run them. The developer published it on GitHub in July 2026 for authorized testing. The FBI reports malicious use since at least August. A month from the initial release to abuse is, for an offensive security tool. After reports of misuse surfaced, the developer pulled the repository on October 8. But, forks, mirrors, archives, and a commercial version keep it in circulation because the internet has never once given anything back. The Internet never forgets. By early November, researchers counted more than 1,600 publicly exposed ARTEX consoles. The reported targets span at least ten market segments worldwide, including financial services, government, healthcare, and energy. One financially motivated campaign exfiltrated personal information on about 67,000 people. In another incident, a supply chain compromise at a cloud backup service opened access to more than 2,250 downstream tenants. The alert maps the activity to 43 ATT&CK techniques. Read it by phase and one thing keeps turning up: a credential somebody forgot about. Reconnaissance. Agents tested the permissions of Azure credentials, including SAS tokens, service principals, Kudu credentials, and storage account keys, with sub-second checks. Initial access. Hardcoded administrator passwords in JavaScript bundles, weak administrative credentials, compromised vendor credentials through trusted B2B federation, and stolen VPN configurations with leaked credentials. Persistence. Account takeover through password resets, new SSH key pairs, and credentials added to existing OAuth applications https://1password.com/blog/protect-against-oauth-supply-chain-breaches . Privilege escalation. Cloud compute instances launched with high-value IAM roles, harvested Kerberos Service Tickets, DCSync to steal the Kerberos Ticket Granting Tickets KRBTGT key, and an administrator token forged with unrotated authentication keys. Internal reconnaissance. More than 110,000 Active Directory objects enumerated, and Local Administrator Password Solution LAPS passwords read. On paper, an admin password in a JavaScript bundle is T1552.001. In practice, it is a sticky note on the monitor, served from a CDN. One piece of good news: endpoint protection blocked attempted lateral movement to domain controllers through Pass-the-Ticket, WinRM, Kerberos relay, and Pass-the-Hash. So boring controls are still earning their keep. Automation changes the price of one exposed credential. A hardcoded password or an unrotated signing key used to wait for a human with time and motivation. Now an agent tests it in under a second and feeds the result into the next technique https://1password.com/blog/ai-credential-harvesting . Two variables decide how quickly an exposed credential can become a broader incident: Standing access sets how far the credential reaches once it is used Detection time sets how long you have, and three hours is less than some incident bridges need to agree on a severity level So we have stopped planning around a patient attacker. Assume anything discoverable gets discovered, and anything valid gets tried. The FBI's credential recommendations are specific. The controls below are theirs. Sweep for embedded secrets. Review application configurations, environment variables, infrastructure-as-code, deployment scripts, and CI/CD pipelines. I would add shipped front-end bundles, since that is where the alert found admin passwords. Rotate what you find, promptly. Include signing and authentication keys. The forged admin token in the alert relied on keys nobody had rotated. Restrict deployment and publishing credentials. List every person and pipeline that can read them today. Cut the list to the ones that deploy. Apply least privilege to multi-tenant application permissions and consent grants. Revoke what nobody uses. Monitor credential additions and privileged role assignments. Monitor the speed of identity access, and spikes in API traffic. Steps one and two fit in a week for most teams. Steps three to five need a named owner and a date, or they become a slide. Several 1Password capabilities are aligned with those recommendations. The FBI does not endorse products, and each capability has a boundary you should know before you count on it. Developer Watchtower https://www.1password.dev/watchtower , in the 1Password desktop apps for Mac, Windows, and Linux, finds unencrypted or insecure SSH keys in ~/.ssh and .env files with plaintext secrets in nine named development folders. It only checks files on local disk that match .env naming patterns, so it doesn’t look at remote repositories or Git history, and other file types like a JavaScript bundle are outside its view even when they sit in a scanned folder.. Privileged Access https://1password.com/product/privileged-access provides just-in-time, task-scoped access to cloud environments with automatic deprovisioning, and discovers overprivileged cloud identities. It applies only to access granted through Privileged Access. SaaS Manager https://1password.com/product/saas-manager finds apps with high-risk OAuth scopes and revokes their tokens in Google Workspace and Microsoft Entra ID. 1Password Developer Tools http://developer.1password.com , allow for JIT secret injection to coding agents and CI/CD pipelines including GitHub Actions https://www.1password.dev/brokered-access/github-actions , validates the workload identity against registered policy and releases only the secrets that are scoped to the task only for the duration of the workflow. Secrets are available to the job during the run. They are not single-use or bound to a runner. None of this is a claim about the ARTEX intrusions themselves. A vendor telling you its product would have stopped a breach it was not present for should make you check your wallet. Most of the ARTEX activity sits outside credential management: web application exploitation, Kerberos attacks, web shells, and command and control. The FBI's guidance also covers reducing the external attack surface, strengthening application protections, and preserving evidence for response. A vault does nothing for an unpatched web app. The facts here reflect the FLASH as published on November 13, 2026. They may change as investigations continue. Read FBI FLASH-20261113-001 https://www.ic3.gov/CSA/2026/261113.pdf and send the mitigations section to the person responsible for your CI/CD pipelines Search your production JavaScript bundles for the string "password." A hit is your first rotation ticket and your opening story at the next all-hands Explore 1Password Developer Tools to learn how to help your team find exposed secrets and keep live credentials out of code and pipelines.