Q4 2023 Policy Summary

Last week, three major events occurred in the AI policy space: the US government's AI Executive Order, the G7's International Code of Conduct, and the AI Safety Summit and Bletchley Declaration held at Bletchley Park in the UK.

The US Executive Order spans over 100 pages and addresses privacy, fairness, bias, and catastrophic risk. In particular, it directs NIST (National Institute of Standards and Technology) to develop model performance and safety evaluations, expand the Secure Software Development Framework (SSDF), and develop a Risk Management Framework (RMF) for generative AI. It also launches a NAIRR (National AI Research Resource) pilot project to improve academic researchers' access to data and computing resources.

The G7 International Code of Conduct was developed through the Hiroshima Process and builds on the White House's voluntary AI company commitments. It sets out responsible practices to mitigate risks throughout the AI development and deployment lifecycle, including evaluation, information sharing, governance, security procedures, and transparency measures. This could become a benchmark for international best practices for frontier AI companies.

The AI Safety Summit held in the UK brought together 28 countries to discuss the risks of frontier AI and response measures, culminating in the Bletchley Declaration. Anthropic's CEO Dario Amodei presented on Anthropic's Responsible Scaling Policy, positioning it as a prototype for a regulatory approach.