Sarah Heck said AI safety cannot depend on informal promises at POLITICO's Decoded summit.
Why it matters: The remarks put Anthropic squarely in the middle of Washington's fight over whether frontier AI should be governed by laws, state standards or voluntary commitments. They also signal what kind of regulation the company is pressing for as lawmakers weigh new rules.
- Sarah Heck, Anthropic's head of public policy, spoke at POLITICO's Decoded summit in Washington, D.C., on Sept. 16, 2026.
- Heck said the U.S. needs to stay ahead in AI to improve safety, rather than rely on an honor code.
- Anthropic says it supports laws with transparency requirements, including California's SB 53, New York's RAISE Act, Illinois' SB 315 and the EU Code of Practice.
- Anthropic's Responsible Scaling Policy says the company uses internal safety and security measures before training or deploying frontier models.
Anthropic's Sarah Heck said at POLITICO's Decoded summit that AI safety cannot depend on informal promises. She said the U.S. needs to stay ahead in AI if it wants to improve safety, according to the summit coverage.
The comments add to a broader policy fight over how frontier AI should be governed in the U.S. and abroad. On its policy page, Anthropic says it backs laws with transparency requirements, including California's SB 53, New York's RAISE Act, Illinois' SB 315 and the EU Code of Practice.
The company also lays out its internal approach in its Responsible Scaling Policy, which says Anthropic uses safety and security measures before training or deploying frontier models. Anthropic last updated that policy on July 8, 2026.
In a prior post, Anthropic argued that federal legislation is the best way to regulate catastrophic AI risks because it would be uniform across the U.S. and could draw on federal expertise in bioterrorism and cybersecurity.
The company also said in a September 2026 threat report that its models had been misused by governments, state-aligned propaganda operations, private firms selling influence, domestic political operators and an opposition movement in exile. That report is part of the company’s public case for more formal oversight.
For policymakers, the open question remains whether AI guardrails should come from federal law, state laws or a mix of both. Heck's remarks put Anthropic on the side of formal standards over voluntary commitments.
By the numbers
- Sept. 16, 2026 - date of Heck's remarks at POLITICO's Decoded summit.
- July 8, 2026 - date Anthropic last updated its Responsible Scaling Policy.
Yes, but: Anthropic's policy page supports formal transparency laws, but the company also relies on its own internal safety policy, underscoring the gap between voluntary practice and public regulation.