Anthropic CEO Says AI Must Slow Down: What U.S. Lawmakers Are Doing About It

CEO’s warning reframes the urgency

Dario Amodei, the chief executive of leading artificial intelligence company Anthropic, made a direct appeal over the weekend: the AI industry should consciously slow down its rapid development so that safety guardrails can be properly designed, tested, and implemented. “The artificial-intelligence industry should slow its fast-moving development to give safety measures time to catch up,” Amodei said on Saturday. The statement instantly reverberated through Washington, where lawmakers have been trying to craft a regulatory framework that can keep pace with an industry that evolves by the month.

“The artificial-intelligence industry should slow its fast-moving development to give safety measures time to catch up.” – Dario Amodei, CEO of Anthropic

Anthropic, known for its Claude family of AI models, has long positioned itself as a safety-first organization. For its CEO to publicly urge restraint—not just for his own lab but for the entire sector—signals that even AI leaders who emphasize caution are worried the current trajectory is outpacing the world’s ability to handle the risks.

Washington’s scramble to build AI guardrails

The U.S. government, however, is not starting from zero. Over the past two years, a mix of congressional hearings, draft legislation, executive orders, and agency guidelines has taken shape. The central challenge is balancing two powerful forces: the desire to lead the global AI revolution—and reap its economic and strategic benefits—against the need to prevent catastrophic errors, rampant disinformation, or a slow erosion of human control.

Congress weighs bills on oversight and accountability

In the House and Senate, lawmakers from both parties have put forward multiple bills designed to increase transparency and impose accountability on AI developers. Among the most discussed are the Algorithmic Accountability Act, which would require companies to assess their systems for bias and safety before deployment, and a bipartisan legislative framework proposed by Senators Richard Blumenthal and Josh Hawley. That framework envisions a new independent federal oversight body with the power to audit, license, and enforce safety standards for the most powerful AI systems.

Lawmakers have also held high-profile hearings in which CEOs of OpenAI, Anthropic, Google, Microsoft, and other major players were questioned on the record about their safety practices, data sourcing, and willingness to accept external regulation. In many sessions, executives agreed in principle that some rules are necessary, but the details of who should regulate, how stringent the rules should be, and what constitutes a high-risk application remain hotly contested.

Executive branch moves ahead with existing authority

While Congress deliberates, the Biden administration has not waited. The centerpiece is the October 2023 executive order on safe, secure, and trustworthy artificial intelligence, which invoked the Defense Production Act for the first time in the AI context. It compels developers of the largest models—those trained on massive computing resources—to share safety test results with the government before releasing them to the public. The order also directed the National Institute of Standards and Technology to finalize its AI Risk Management Framework, a practical guidebook for organizations to identify, assess, and mitigate AI-related threats across the entire lifecycle of a system.

Other agencies, from the Federal Trade Commission to the Department of Commerce, have opened inquiries into how existing laws on consumer protection and civil rights apply to AI, while the White House has secured voluntary commitments from 15 leading companies to invest in safety testing and watermarking of AI-generated content.

A multi-pronged push

Key regulatory efforts now underway include:

• The White House executive order mandating safety test disclosure for the most powerful models.
• NIST’s AI Risk Management Framework, a voluntary yet influential standard.
• Proposed legislation such as the Algorithmic Accountability Act and the bipartisan Blumenthal-Hawley oversight framework.
• FTC and Commerce Department investigations into how existing laws apply to AI.
• Voluntary safety commitments signed by 15 leading AI firms, including Anthropic.

The overlapping actions reflect both the urgency Washington feels and the lack of a single consensus on how to proceed.

A patchwork approach taking shape

The result is a regulatory environment that currently looks more like a mosaic than a single, coherent law. That patchwork—voluntary pledges, binding executive orders, agency guidance, and proposed bills—has its critics. Some industry advocates warn that overlapping, sometimes conflicting rules could stifle innovation and push development offshore. Consumer and civil-society groups argue that voluntary measures lack teeth and that Congress must pass a clear statute with enforceable penalties.

What regulation could mean in practice

For the companies building frontier AI models, any final regulatory package would likely mean new obligations: documenting data sources, proving that models have been tested for harmful outputs, submitting to external audits, and possibly obtaining a license before deploying particularly powerful systems. For businesses that use AI—from healthcare diagnostics to hiring algorithms—the rules could require them to maintain detailed records and provide explanations for automated decisions. For ordinary consumers, regulation could mean stronger protections against biased lending, misleading deepfakes, and privacy invasions, along with clearer rights to opt out of or contest AI-driven outcomes.

The question of pace

Amodei’s call to slow down is not universally shared inside the industry. Some competitors argue that a voluntary slowdown is unrealistic in a competitive market and that properly designed regulation, not a moratorium, is the right answer. Yet his warning adds an influential voice to the argument that government must move faster than it has. With Congress back in session and multiple committees drafting AI bills, there is a chance that the coming months bring committee votes and possibly floor action. Whether that leads to a signed law by the end of the year remains uncertain—but the pressure from inside the AI labs themselves is making the status quo harder to defend.