1Password’s Off-by-1 Labs reports that frontier LLMs fully fixed six recent open-source vulnerabilities without behavior changes only 26% of the time, releasing FLAWED tooling and warning that AI-generated security patches still need expert human review.
Off-by-1 Labs Research: AI-generated vulnerability patches require human review
1Password's Off-by-1 Labs reports that frontier LLMs fully fixed six recent open-source vulnerabilities without behavior changes only 26% of the time, releasing FLAWED tooling and warning that AI-generated security patches still need expert human review.
Source: 1Password