Study: coding agents propagate existing vulnerabilities into new code they write
OfirPress · x · 2026-11-10
New work shared by Ofir Press shows coding agents propagate vulnerabilities: when a repo already contains vulnerable code, agents write similar flaws into new code—even when the vulnerable code is functionally unrelated to their task and the agent can detect the flaw. Implications for code review around AI-written code.
Related event: Coding Agents Spread Security Vulnerabilities From Codebases, Study Finds(2 posts)→
More from coding & agent
- Fable 5.2 as orchestrator paired with Opus 5.5 as executor is 'breaking my brain' — talkaboutdesign · 2026-11-10
- Anthropic engineer's AI-built browser FPS nears COD quality, sparking 'one year to Sea of Thieves' claim — ChrisGPT · 2026-11-10
- Tech debt in the AI era: paid in full whenever the next frontier model ships — doodlestein · 2026-11-10
- Claude-coded Valheim mod rebuilds Ocarina of Time's Kokiri Forest from your own ROM — Any-Explanation3814 · 2026-11-10
- App Store Connect CLI 6.5 adds subscription retention messages and offers — rudrank · 2026-11-10
- App Store Connect CLI 6.6.0 ships nearly 50 bug fixes, adds Nix flake and no-wait screenshot uploads — rudrank · 2026-11-10