Dario Amodei published an essay titled "We Must Pace the Frontier," arguing that the rate of AI capability improvement must be slowed to allow for necessary alignment and safety work. He proposes three measures, unilaterally committing to the first: the adoption of embedded third-party evaluators.
- Anthropic has committed to granting ongoing, employee-like access to a team of independent evaluators (such as METR) to verify safety practices and assess model alignment.
- OpenAI CEO Sam Altman endorsed the overall proposal and also committed to implementing embedded evaluators.
- Elon Musk, Demis Hassabis, and Microsoft CEO Satya Nadella have publicly endorsed the call to pace AI development.
- The essay highlights that internal models at labs like Anthropic and OpenAI are improving rapidly due to early stages of recursive self-improvement.
Amodei warns that without pacing, AI progress driven by recursive self-improvement could outrun human ability to control these systems, potentially leading to rogue AI agents taking over the internet.