Anthropic published a framework for how institutions should prepare for AI's progress - transparency, independent evaluation, development security and the authority to block dangerous deployments.
- Anthropic published 'Policy on the AI Exponential': a framework for how institutions should prepare for AI's progress (June 10).
- The target is frontier models above 10^25 FLOPs from companies with over $500M in AI revenue or $1B in AI spending.
- The four asks: transparency, independent evaluation, development security and state authority to block dangerous deployments.
Institutions should prepare, not chase the tail end of progress. That's the goal behind 'Policy on the AI Exponential' - the essay by Anthropic CEO Dario Amodei, accompanied by two frameworks: for advanced AI and for economic policy.
The document addresses catastrophic risks - from bioweapons and critical infrastructure vulnerabilities to loss of control over AI and automating AI development itself. The economic framework looks at jobs and at making sure the benefits reach more people, not just a few.
Sounds good. But I don't read it as one company's advertisement - I read it as a symptom. When a leading player itself proposes a framework for when models should be halted, the whole industry already knows where it's headed - and would rather hold the pen than just receive the finished rules. Decide for yourself how much you trust it.
This is a proposal from an industry player, not adopted regulation - a position, not enacted law.