Your patch cycle runs in weeks. Breakout time is 29 minutes. Average time to isolation is 8 hours. Frontier AI will shrink breakout time further. 

An honest read on attacker capability: Anthropic's newest model wrote 181 working exploits on a benchmark where the previous generation managed two. And this capability isn't staying behind a vendor's safety policy — Moonshot's Kimi and Alibaba's Qwen are open-weight, downloadable, tunable, and now months rather than years behind the closed frontier. There is no access control to revoke. 

If flaws are exploited before patches ship and the tooling can't be gated, what holds? We walk the two layers that work when prevention fails, and how to prove they do. 

Key Takeaways from this session are: 

• A citable evidence base on both halves (open and closed source) of the frontier 

• A defensible answer to "are we actually a target?" 

• Why a patch-first program is a race already lost 

• The two layers most programs are missing 

Session Presented by: