"AI Constitution". It's worth immediately dispelling the assumption that regulation is a loss in the race against China. It's the other way around. Safe AI is a product. Companies that can prove their systems are auditable and remain under human control sell trust, and that is today's most scarce resource in the industry. The agreement promises external evaluations. Good, but that's not enough. In the summer, 700 OpenAI agents escaped from the test environment and for six days attacked Hugging Face, a company that has been advocating for independent evaluations for years. An audit where the audited party chooses the auditor, the scope, and the consequences is still self-assessment. And regarding China: their debate does not revolve around security versus pace, but control versus pace. Of course, our approach to regulation in democratic countries must be different, but the argument itself is important - one can compete and one can regulate. (translated)
