Cognition has launched Devin Security Threats, a managed catalogue of newly disclosed vulnerabilities that connects directly to its Devin coding agent. Instead of leaving teams to chase each new CVE across their repositories, every threat entry becomes a starting point for a scan, a triage queue and a prepared fix.
The launch arrives alongside two serious disclosures. Java4Shell is a Java runtime flaw that can expose memory or allow remote code execution, tracked as CVE-2026-114938. NGINX2Root is a group of five memory-corruption flaws in nginx with a similar remote code execution risk. Both underpin much of production software.
From a threat card, a developer can start a scan across repositories connected to Devin. The system splits the investigation across parallel agent sessions using the same Agentic MapReduce process as the rest of Devin Security, then consolidates the results into one set of findings. Teams can also switch on automatic triage: pick repositories and a minimum severity, and Devin scans them whenever a qualifying threat is published. Findings are ranked against the team’s own threat model, and threats with no findings can be closed automatically to keep the queue clean.
Fixes stop short of deployment: Devin prepares code changes as pull requests for review, testing and approval. Internal teams can also create their own threat records for private findings such as bug-bounty reports, so the workflow covers proprietary code patterns as well as public CVEs. Cognition has also published two open-source skills for the new threats, fix-nginx-vuln and fix-java-vuln, which teams can inspect, adapt as internal checklists, or hand to other coding agents such as Cursor and GitHub Copilot. Pricing and supported repository hosts have not been disclosed, so teams will need to judge scan accuracy and patch quality on their own codebases.
Source: AlphaSignal