The AI regulation smackdown isn’t over

Anthropic chief executive Dario Amodei unveiled a three‑step plan to slow the pace of artificial‑intelligence development on Monday at a closed‑door industry summit in San Francisco, calling for third‑party evaluators inside labs, coordinated domestic standards and a framework for international accords.

The proposal comes after a wave of high‑profile AI releases and safety scares that have put regulators on high alert. In November, OpenAI rolled out GPT‑4, prompting lawmakers in Washington and Brussels to demand more oversight. The U.S. Senate held its first hearing on AI risks in February, and the European Union pushed forward its AI Act, which aims to classify and limit the most dangerous systems. Industry leaders have been divided, with some urging rapid innovation while others warn that unchecked progress could outpace safety measures.

Amodei’s plan echoes calls from the White House’s Office of Science and Technology Policy for “shared responsibility” among developers. He suggested that independent auditors be granted access to model training data and code, a move praised by consumer‑advocacy groups but met with resistance from firms fearing trade‑secret loss. The domestic coordination step would create a voluntary code of conduct for U.S. labs, similar to the “AI Safety Partnership” formed earlier this year. Internationally, Amodei urged the U.K., EU, Japan and the U.S. to negotiate binding limits on model size and compute power, a notion that has already sparked debate in the G‑7 AI summit slated for next month.

If the plan gains traction, it could reshape how AI companies operate and influence forthcoming legislation. Congress is expected to draft a bipartisan bill that mirrors parts of Amodei’s framework, while the European Commission is reviewing the feasibility of cross‑border evaluator standards. Critics warn the approach may slow beneficial AI applications, but supporters argue that a measured pace is essential to avoid catastrophic outcomes. The next weeks will test whether the industry can rally around a shared set of rules or remain fragmented in its pursuit of ever‑more powerful models.