AI Safety
Only 26% of AI Security Patches Were Clean. Another 20% Fixed the Bug and Broke the App.
1Password's research lab graded 6,080 model-written patches against six recent CVEs. More than a third of the successful ones were rated fragile.
2h ago
