Anthropic policy chief rejects “honor code” for AI oversight

Sarah Heck, head of public policy at Anthropic, said AI companies cannot rely on an internal “honor code” to manage safety and oversight. She called for closer coordination with government as regulators and industry leaders intensify debates over how fast AI models should advance.

No “checking our own homework”

Speaking at the Politico Decoded Summit in Washington on Wednesday, Heck said oversight cannot be self-administered. “We can’t be checking our own homework, and that’s very clear,” she said, adding that Anthropic wants to work with government to determine what standards make sense.

CEO Amodei urges deliberate slowdown

Heck’s comments followed days after Anthropic CEO Dario Amodei urged the industry to deliberately slow the pace of model development. Amodei proposed a three-step plan on Saturday aimed at tempering how quickly model capabilities improve without giving up “commercial advantage or the United States’ lead in AI.”

Mixed industry reaction

Several executives backed Amodei’s stance. Elon Musk wrote “Dario is right” on X on Saturday. Other leaders including Nvidia CEO Jensen Huang and Meta CEO Mark Zuckerberg were less convinced that development should be intentionally slowed.

At Salesforce’s Dreamforce conference on Tuesday, Huang said the industry does not need new laws or regulations and that speed and safety are not mutually exclusive. Zuckerberg echoed that view on Tuesday and said Meta delayed the release of Muse, its personal AI agent, for “several months” to focus on safety and security. Zuckerberg wrote on X that Meta “didn’t call for everyone else to do this before we would.”

Ongoing talks with White House and Congress

Heck said Anthropic speaks with the White House on a “daily basis” and has been working with Congress “all year.” She said the company is trying to ask “the hard questions now,” including how to handle China, how to regulate AI, and how to advance safeguards before they become harder to implement.

Why it matters

The exchange adds pressure to the central policy fight in AI: whether safety can be enforced through voluntary industry standards or requires external oversight as model capabilities accelerate.