Cognition says artificial intelligence is rapidly increasing the volume of newly discovered software vulnerabilities. According to Cognition's blog post, two major disclosures in foundational software arrived in the same week. Java4Shell, published November 17, is a Java runtime vulnerability that can lead to memory disclosure and remote code execution, Cognition says. NGINX2Root, published November 18, consists of five memory-corruption bugs in nginx that can also lead to remote code execution, according to the post.1

In Cognition's model, each threat describes what is affected, how an attacker would exploit it, and what fixes the issue. Selecting Scan with Devin opens a new scan with that threat preselected under Focus on threats, according to the blog post. Teams pick the repositories to inspect, and Devin runs the same Agentic MapReduce process used in standard Devin Security scans, Cognition says. Findings are then ranked against the team's threat model, and developers can assign them to Devin to open pull requests for review.1

Organizations are not restricted to entries published by Cognition, according to the announcement. If a bug bounty report, vendor notice, or incident review raises a concern about internal code, a team can define a custom threat with a title and description and scan for it the same way, Cognition says. For teams using Auto-triage, Cognition notes that a threat whose scan finds nothing can be dismissed automatically.1

Both initial threats are live at app.devin.ai/security, Cognition says. For teams that prefer running remediation inside a Devin session rather than a scan, the Vulnerability Fixes plugin adds the fix-java-vuln skill for Java4Shell and the fix-nginx-vuln skill for NGINX2Root, according to the post. Developers can install the plugin from the Devin Marketplace or via the Devin CLI, and Cognition says the plain Markdown skills can also be read as a checklist or given to another coding agent.1

Cognition's announcement does not specify pricing, token usage, or scan duration for running threat scans or remediation sessions. It also leaves review of generated pull requests to the engineering team, according to the post.1