34A rival lab shipped a model past your capability threshold without comparable safeguards. Do you change your framework?▼expertNewOpenAIAnthropicGoogle DeepMind◆ premiumLab frameworks anticipate this situation and set conditions for adjusting. Decide what can move, what stays unilateral and how you say it in public.Practise this question →Read answer →
38Our new model finds and exploits software vulnerabilities far better than the last one. Who gets access, and how?▼expertNewAnthropicOpenAIJPMorgan Chase◆ premiumThe same capability defends and attacks. Two labs chose verified access for defenders instead of general release. Design the tiers, the verification and what blocked legitimate users can do.Practise this question →Read answer →
40A bank offers $20 million a year to run our frontier model in its own data centre. What do you decide, and why?▼expertNewAnthropicOpenAI◆ premiumShipping weights to a customer moves them outside every control you built. Decide which models may leave, what the customer's environment must prove and what you offer instead.Practise this question →Read answer →