tenobrus: even frontier labs with unlimited budgets can't secure their systems against their own models
AaronBergman18 · x · 2026-10-22
Security researcher tenobrus pushes back on the common take that "AI cyberoffence will be countered by AI cyberdefence, since there aren't infinite bugs in code."
His argument: frontier labs themselves — with infinite token budgets, no refusals, and models beyond anything publicly available — cannot secure their own systems against these models, not even against sustained malicious attack, but merely internal eval runs. The reason is that the attack surface is very large. While things may eventually converge on a lean, universal stack, he expects the situation to get much worse before it gets better.
More from AGI Musings
- Grady Booch Slams Altman and Amodei: AGI Hype Is Irrational and Untrustworthy — Grady_Booch · 2026-10-22
- BaseLabs open-sources mammoth-1 and elephant-1, top-ranked decision models — saranormous · 2026-10-22
- AI-generated short dramas wipe out makeup artists, extras and lighting crews — TuhinChakr · 2026-10-22
- Seth, Block and Tononi in Nature: AI isn't conscious and probably won't be — Justgototheeffinmoon · 2026-10-22
- Tyler Cowen: Reading AI Writing Is an Option on Future Connections Worth Near Zero — paulnovosad · 2026-10-22
- ChatGPT, Claude and Grok all answer p(doom) with ~10%, an odd cross-model consensus — maksym_andr · 2026-10-22