Autonomous agents convert AI-written security patches into working exploits without human assistance Researchers found that an autonomous AI agent converted a security patch co-written by Anthropic's Claude for the vm2 JavaScript sandboxing library into a working exploit without human assistance, according to CSO Online and The Register. The vm2 library is downloaded 1 million times weekly, and the incident shows how AI assistance on both the defensive and offensive sides can turn patches into attack maps. Autonomous agents convert AI-written security patches into working exploits without human assistance According to CSO Online and The Register, researchers found that attackers used an AI agent to convert security patches co-written by Anthropic's Claude into working exploits. The vm2 JavaScript sandboxing library, downloaded 1 million times weekly, shipped an AI-assisted patch that was subsequently reversed into an attack by an autonomous agent. The incident demonstrates a vulnerability in the patch-and-exploit cycle when AI assistance is involved in both security defenses and attack discovery. Topics Sources - Press Read article https://www.csoonline.com/article/4248687/ai-written-patches-are-becoming-attack-maps-for-hackers.html - Press Read article https://www.theregister.com/research/2026/11/16/google-ai-agent-turned-a-security-patch-into-a-working-exploit-with-no-human-help/5309926 Go deeper This intelligence is sourced automatically from public sources across the web and synthesised by the Prefactor AI pipeline. Stories are reviewed before publication.