AI-written patches are becoming attack maps for hackers
AI-generated summary Anthropic’s Claude co-wrote 19 fixes for vm2 sandbox-escape vulnerabilities, while Rubrik lead product security engineer Shayaun Nejad warned that patches can serve as blueprints for attackers. OpenAI security researcher Tim de Waal and researchers at 1Password’s Off-by-1 Labs demonstrated that AI-generated patches frequently leave attack paths open.