1/7 Increasing automation of R&D processes using frontier models is both enabling progress in critical fields like biomedical research and mathematics, and opening the door to severe threats, some known national security risks, others hard to predict as AI capabilities advance.

6/7 Visibility, control and verification of automated research agents, and long-lead federal capacity should all be addressed now if we are to maximize the benefits of automated R&D while avoiding its most dangerous pitfalls.

1/3 As capabilities of AI models accelerate, comprehensive policies to govern the most powerful systems, advance American AI leadership, and strengthen resilience against AI-enabled threats are urgently required.

2/3 The forward-looking policy agenda released today by @iapsAI sets out how to drive AI development, preserving American innovation while introducing safeguards for the most serious harms.

1/3 Apply now for the 2027 Spring IAPS AI Policy Fellowship: a fully funded, three-month Fellowship for professionals looking to pivot into AI policy and help meet the magnitude of challenges emerging from the frontier.

2/3 Whether your background is in policy, public service, tech, government, law, academia, or another field, the @iapsAI Fellowship provides the skills, mentorship, and network to help you make an impact.

1/7 With the Trump-Xi summit set to get underway in late September, U.S. and Chinese officials plan to convene for a dialogue on shared risks from AI.

6/7 Our proposed U.S.-China AI Risk and Incident Dialogue would serve four functions: development of shared definitions for AI risks and incidents; routine sharing of non-sensitive information; establishment of AI-incident notification procedures; and post-incident consultation.