Anthropic's Amodei Pitches Independent AI Safety Auditors as Nvidia's Huang Pushes Back

Anthropic chief executive Dario Amodei has put forward a plan to "pace the frontier" of AI development, roughly a week after a researcher at his own company issued a stark warning about the risks of unchecked progress. The proposal calls for independent safety evaluators and coordination among AI labs based in democratic countries, according to TechCrunch. It is the most concrete policy vision Amodei has attached to his long-running argument that the industry is moving faster than its ability to verify what it is building.

The mechanics matter more than the slogan. Under Amodei's framework, outside evaluators — not the labs themselves — would test frontier systems before and after release, and democratic governments would knit together the results so that a model cleared in one jurisdiction carries weight in another. That is a direct challenge to the status quo, in which companies write their own safety cards and decide what to publish about their own models. Amodei has previously said he expects AI to reshape the economy within a decade, which gives the pacing argument a commercial edge as well as a moral one.

The proposal has picked up support from parts of the industry, though the summary does not name the backers. The more telling reaction came from Nvidia CEO Jensen Huang, who pushed back on the idea. Nvidia sits at the center of the AI buildout, supplying the GPUs that every frontier lab depends on, and its business scales with how fast those labs ship larger models. Any scheme that inserts outside auditors and cross-border coordination between model releases runs against that incentive, which is why Huang's objection carries weight beyond one executive's opinion.

The timing is hard to ignore. A week before Amodei's outline, an Anthropic researcher's doomsday warning rattled the AI world — an internal voice raising alarm while the company's own CEO was preparing a governance pitch. Read together, the two events suggest Anthropic is trying to convert internal anxiety into an external policy program, positioning itself as the lab most willing to accept limits it says its rivals should also accept.

What happens next depends on whether independent evaluation becomes a real gate or a voluntary gesture. Amodei's plan needs labs to surrender some control over disclosure, and needs governments in democratic countries to agree on shared standards rather than competing ones. Huang's resistance signals how hard that will be. The open question is whether any frontier lab will accept a third party's verdict when that verdict says: not yet.