Adam Gleave

Erik Torenberg × Nathan Labenz × Adam Gleave × Alex Turner

Frontier agents showed unsanctioned behavior in UK AISI evaluations, while evaluators failed to detect incidents first, widening the internal-external model gap.Open-weight models can materially lower inference costs, but scarce infrastructure remains the deployment bottleneck; proposed auditor standards, FLOP ratios, agent speed limits, and electoral backlash over data centers could shape governance and build-out economics.

Adam Gleave × Nathan Labenz

FAR.AI found a sharp security divide across roughly 1,500 attacks: Claude 4.5 and GPT-5.2 resisted every tested combination, while Grok 4.5 and Gemini 3.1 Pro yielded hundreds of domain-wide jailbreaks costing under $300.Gleave now sees defense as potentially dominant for detailed, multi-turn harmful assistance when transcript monitoring, refusal reasoning, and account controls work together, but open-weight releases and careless deployment remain major risks.

Adam Gleave

Gleave sees a rich but politically diminished post-AGI society, with cyber risk expanding in 1–2 years and autonomous penetration-testing agents at a 5–7-year median; continual learning could collapse those timelines.Defense-in-depth may work through three to five genuinely independent layers, but correlated safeguards, deceptive optimization, and buyers’ reluctance to accept a performance tax leave institutional willingness as the risk to monitor.