CodeRabbit went looking for trouble in open-source code and isn’t done yet

CodeRabbit's Secure Deep Scans identified 1,215 high or critical security findings across 1,125 public repositories between October 23 and November 3, of which at least 75 have been fixed, with fixes offered to maintainers at no cost. The campaign follows a wave of open-source security incidents, including PHP's October 21 emergency releases for two critical vulnerabilities, a compromised LiteLLM release that reached PyPI and Docker carrying malware that stole cloud credentials, private keys and AI tool logins, and a Brotli memory bug tracked as CVE-2026-111497 patched that same week by Node.js, Bun and .NET. Herrington Darkholme, creator of ast-grep, said of the tool: "Proving vulnerabilities before flagging them makes every finding far more actionable.

CodeRabbit Secure Deep Scans found more than 1,200 vulnerabilities in open-source repos and are helping fix them at no cost to the maintainers If you maintain an open-source project, you’re now up against attackers who can use AI agents to find and exploit vulnerabilities in your code. Those agents can try different attack paths https://cloud.google.com/blog/topics/threat-intelligence/artex-and-friends-enter-agentic-offensive-frameworks , adapt when something fails, and complete parts of an intrusion in minutes or even seconds. That lets attackers probe more weaknesses faster, putting pressure on maintainers to get fixes out sooner. Recent incidents show how involved that work can become. On October 21, PHP shipped emergency releases for two critical vulnerabilities https://thephp.foundation/blog/2026/10/21/todays-php-security-release-and-how-were-strengthening-php-security/ that allowed unauthenticated attackers to execute code on some PHP servers. The PHP Foundation describes how compatibility trade-offs, limited resources, and difficult decisions repeatedly stalled the fix over the preceding months. The same day, a compromised LiteLLM release https://socket.dev/blog/litellm-compromise reached PyPI and Docker carrying malware that stole cloud credentials, private keys, and AI tool logins. LiteLLM’s root cause analysis https://docs.litellm.ai/blog/security-incident-oct-21-2026-attack-chain traced the attack to an administrative key that had remained valid after being exposed in a public repository in May. That same week, Node.js, Bun, and .NET were patching the same Brotli memory bug https://socket.dev/blog/emudhra-brotli-security-fix , CVE-2026-111497, in their bundled copies of the library. Finding a hidden vulnerability has never been easy. Maintainers need to find them before attackers do, then develop, test, and release fixes that protect users without breaking existing functionality. Traditional rule-based checks remain useful, but they cannot replace investigating how code paths interact and whether the application enforces the protections it needs. That’s where CodeRabbit Secure Deep Scan https://docs.coderabbit.ai/security helps. It examines committed code across a repository, maps the attack surface, and investigates potential vulnerabilities across code paths. A separate verifier checks the supporting evidence, reachability, and existing safeguards. For supported findings, Fix with AI prepares a patch that maintainers can review, test, and merge, helping them move from investigating a report to fixing the underlying problem. What the scans found Between October 23 and November 3, Deep Scans across 1,125 public repositories identified 1,215 high or critical security findings : 1,200 high and 15 critical. At least 75 findings have been fixed. One square represents one finding. Green marks the 75 findings fixed; gray represents the other findings in the campaign total. Herrington Darkholme, creator of ast-grep https://github.com/ast-grep/ast-grep , described his experience: I tried Deep Scan against ast-grep and was impressed by its precision. Proving vulnerabilities before flagging them makes every finding far more actionable. — Herrington Darkholme, creator of ast-grep A critical SQL-injection finding in an intentionally vulnerable OWASP demo, with reachability and exploitability details. What CodeRabbit Secure does Deep Scan https://docs.coderabbit.ai/security investigates the code and infrastructure configuration across an OSS repository, including legacy code. Its workflow has four stages: 1. Map. Identify security-relevant entry points, trust boundaries, access controls, sensitive operations, and configuration. 2. Hunt. Follow the relevant code paths to investigate suspected vulnerabilities, including missing checks and assumptions spread across components. 3. Verify. A separate verifier reopens the cited source, checks reachability and existing safeguards, and challenges the conditions required for exploitation. 4. Fix. For supported findings, generate a proposed patch for a developer to review. Results appear with severity, supporting code, and available remediation actions. Incremental delta scans can revisit changed files and stale or unverified areas after the initial full scan. Attack surface map for the intentionally vulnerable OWASP demo repository. Deep Scan complements rule-based static analysis Static analysis can trace data across functions and identify unsafe flows. Mature tools also model frameworks and complex application behavior. The challenge is accounting for application-specific security requirements that aren’t fully captured by the configured checks. AI agents can help attackers look for ways around existing safeguards, and we need to evolve how we protect our code against them. Consider an endpoint that authenticates a user before returning a document. Authentication establishes who made the request. But the application must also establish whether that user is allowed to access that particular document. The ownership check might live in the route, a service, or the database query. Deep Scan investigates those relationships. It looks for the control, follows the reachable request path, and checks whether another safeguard invalidates the suspected vulnerability. Each reported finding includes supporting code and reasoning so maintainers can inspect the conclusion. We explored this approach in Why it’s time to fight AI with AI https://www.coderabbit.ai/blog/fighting-ai-with-ai . Deep Scan adds to the security tools you already use, investigating application-specific flaws that may fall outside their configured checks. Static analysis and dependency scanning should remain part of your workflow. Security blast radius shows code connections and potentially impacted components in an example pull request. Most common security vulnerabilities The most common vulnerabilities Deep Scans found in open-source repos include: - Missing ownership checks: An endpoint checks that a user is signed in but not that they are allowed to see or change the specific record. It is easy to miss in review because the auth middleware is right there in the route. - Unsafe browser script handling: Untrusted input reaches the page without being escaped as the surrounding code expects, allowing cross-site scripting. - Weak identity checks: Tokens, signatures, or callers accepted without fully verifying who sent them. - Unsafe file path handling: User-supplied paths or archive entries that can escape their intended directory. - Incomplete session expiry checks: Sessions or tokens that stay valid after they should have expired or been revoked. - Information exposure: Responses, errors, or storage that reveal data to someone who should not see it. From finding to a fix When Deep Scan reports a vulnerability, the finding includes the affected code, the reasoning behind it, and a recommendation for fixing it. Maintainers can use that evidence to begin work on a patch without reconstructing the investigation. For supported findings, Fix with AI https://docs.coderabbit.ai/security fix-with-ai sends the finding to CodeRabbit Agent, which prepares a patch and opens a pull request for review. You can follow the task, ask for changes, and review the diff. If you prefer another coding agent, select Share → For agents https://docs.coderabbit.ai/security filter-share-and-export to copy the finding details and recommendation as self-contained Markdown. Paste that into the agent you already use and ask it to implement the fix in your repository. Our $1 million free Deep Scan promotion has ended, but our commitment of more than $10 million to open source https://www.coderabbit.ai/blog/coderabbit-expands-its-commitment-to-open-source continues. We’re offering one free Deep Scan to open-source projects trying it for the first time. The offer covers up to $1,000 for the scan, and projects must have two or more members to qualify. Your first Deep Scan is on us. Get started at app.coderabbit.ai/security https://app.coderabbit.ai/security or reach out to oss@coderabbit.ai mailto:oss@coderabbit.ai .