Amodei's Call to Pace the Frontier Tests AI Labs on Safety and Coordination
Anthropic CEO Dario Amodei has proposed to "pace the frontier" of AI development, setting out the idea in a blog post organized around three broad strategies. TechCrunch
The proposal centers on external checks and coordination. It relies on independent safety evaluators to assess frontier systems, and it calls for coordination between AI labs in democratic countries. TechCrunch
OpenAI CEO Sam Altman agreed with Amodei that it is time to "pace the frontier." TechCrunch The proposal has received some industry support more broadly, while also drawing pushback from Jensen Huang. TechCrunch Huang is the chief executive of AI chip maker Nvidia. The New York Times
The public exchange sits alongside quieter discussion among labs. OpenAI, Anthropic and Google have been in talks on AI safety for weeks. TechCrunch Google DeepMind has launched an institute to widen the AGI debate. TechCrunch
On TechCrunch Equity, hosts Kirsten Korosec, Anthony Ha and Sean O'Kane took up the practical question behind the slogan. They discussed whether companies can agree on what slowing down means and who gets to police it. TechCrunch That is the easy part to state. Agreement is harder.
The broader context here is familiar to anyone who has worked around standards, audits and shared infrastructure. Independent evaluation only works if evaluators have access, methods and time. Access means model weights, scaffolding, tool use and deployment context, not just an API endpoint for a few days. Methods mean repeatable tests for capabilities and safeguards that survive small changes in prompting and post-training. Time means evaluation before deployment decisions lock in, when a negative finding can still change a release.
In my view, the coordination question is even more difficult. Labs in democratic countries still compete on talent, compute, enterprise contracts and consumer distribution. A voluntary pace agreement asks competitors to share safety findings, accept external judgment and slow releases that might otherwise win customers. Worth flagging as an open issue is verification. Without a way to confirm what was trained, evaluated and shipped, coordination depends on trust among direct rivals.
Looking at what this means for practitioners, the near-term stakes are procedural rather than technical. If independent evaluators become a normal gate, frontier teams will need evaluation-ready builds, documentation and incident response plans as routine artifacts. If lab-to-lab coordination holds, safety talks of the kind reported among OpenAI, Anthropic and Google could become a standing forum rather than an ad hoc response. If it does not hold, each lab will continue to set its own bar and ask the public to trust internal testing.
My own expectation, after watching my children grow up alongside each platform shift from desktop to mobile to cloud, is that adoption will not wait for governance to feel settled. Developers will build on whatever frontier models are available. That makes the plumbing matter. Clear evaluation criteria, published results where possible, and shared language for risk would let downstream builders make better decisions even while the larger debate over pace continues. The proposal does not settle that debate. It gives it a structure to work with.
