AnalysisCybersecurityJuly 24, 2026
Training Frontier Models to Out-Think Hackers

Talk covers using a frontier model to autonomously exploit a chain of Keycloak, Vault, and broker to reach production code, discovering a genuine zero-day. Speakers Uri Rolls and Thom Wolf discuss training models to out-think hackers.
Featured · Uri Rolls, Thom Wolf
1 source
More stories today
- Seeing AI Agents Is Not Enough. Security Teams Must Enforce What They Can Do
- Proposed 'Genie Coefficient' measures AI alignment gap
- Stripe in talks to acquire OpenRouter for $10 billion
- BrainAPI converts raw text to knowledge graph via agent swarm
- Israel, UK name AI ministers to compete with US, China