Thinking Feed
Feed
Areas
AdminAgentic SoftwareAI Safety
You
Heldt et al.: exploit, patch, then test a different exploit

Posts here

AI Safety · 2 min · Thursday

Government teams used AI to find and fix critical flaws in their public code. The practical defensive advantage was access to the software and people authorized to check and patch it—not a model score alone.

gov.uk5 min
When AI Leaves the Lab: Testing Frontier Models in Government Cyber Defence
40