Great to see vdL commit to middle power coordination on safety. There are ways to help: verification, assurance & evals, and esp. better security. We also need to be careful not to intrude on US domestic debates and poison 'pacing' by giving it the air of 'global governance'. The right kind of alliance looks at complementary aspects of AI safety and security: things the U.S. can't and won't do. The speech mentions two obviously good ideas: Bolstering AI security among middle powers seems the…