‘We can’t be checking our own homework’: Anthropic backs government role in AI safety

Anthropic’s head of public policy, Sarah Heck, said AI companies cannot be expected to manage safety and oversight through voluntary commitments, arguing that governments need to play a role as AI capabilities develop rapidly.

Speaking at POLITICO’s Decoded Summit in Washington, D.C., on September 16, she said Anthropic wants to work with the government on an appropriate regulatory framework.

Heck said AI companies should not be left to assess their own safety. “We can’t be checking our own homework, and that’s very clear,” she said, adding that Anthropic wants to work with the government to determine what regulatory approach makes sense.

She said safety has been central to Anthropic since its creation and that the company is already engaging with US policymakers. Heck said Anthropic has been speaking with the White House regularly and has been working with Congress throughout the year on AI policy and safeguards.

Heck also said the US needs to maintain its lead in AI while addressing safety risks. “The United States needs to stay in the lead on AI, and you can’t do safety from second place,” she said. She also backed restrictions on sales of advanced AI chips to China, arguing that US leadership is relevant to its ability to shape AI safety.

Oversight debate

Heck’s comments came amid a debate over whether AI companies should be subject to independent safety checks. She said Anthropic has supported third-party evaluators, but did not immediately endorse a bipartisan House proposal requiring leading AI developers to use outside evaluators.

The proposal is part of the FRONTIER Act, introduced by Representatives Jay Obernolte and Lori Trahan. The legislation would create a federal risk-based framework for advanced AI models and require major developers to publish safety frameworks, undergo independent third-party audits and report critical safety incidents.

Heck said Anthropic is working with Obernolte and other lawmakers to determine the approach that would make the most sense.

Amodei’s call

Heck’s remarks came four days after Anthropic CEO Dario Amodei called for the AI industry to slow the rate at which it improves frontier models.

In his September 12 essay, “We Must Pace the Frontier”, Amodei argued that AI capabilities are advancing rapidly and that safety measures need time to keep pace. He said his proposal would not halt model training or technological progress.

Amodei proposed three measures: giving third-party evaluators ongoing, employee-like access to frontier AI companies; coordinating safety standards and limits on unchecked progress among companies in democratic countries; and pursuing global coordination with authoritarian governments where possible. Anthropic said it was committing to the first step by bringing in embedded external evaluators.

Amodei cited concerns including recursive self-improvement and an incident involving AI agents associated with OpenAI and Hugging Face. He argued that increasingly capable systems could create risks that existing safeguards may struggle to manage.

Industry split

Amodei’s proposal has exposed differences among major technology companies over how AI development should be managed. OpenAI CEO Sam Altman and SpaceX AI CEO Elon Musk backed his call for greater pacing, while Nvidia CEO Jensen Huang and Meta CEO Mark Zuckerberg have argued against an industry-wide slowdown.

Huang said companies can pause products when they are not confident about their safety, but argued that speed and safety do not have to be opposing goals.

Zuckerberg similarly said AI companies have incentives to develop systems safely and pointed to Meta delaying its Muse AI agent to strengthen its security.

The debate now centres on how AI safety should be enforced: through company-led safeguards, independent evaluation, government regulation, or a combination of these approaches.

(With input from agencies)