Regulation Policy

Anthropic’s AI Safety Bill Could Limit Local Protections

A tech giant’s push to shape AI laws may weaken government oversight and reduce community control over how AI impacts daily life.

Published October 17, 2026 Read 4 min 1354 words By Jonathan West Via Politico ↗

Anthropic, one of the world’s largest artificial intelligence companies, is quietly lobbying Congress to pass an AI safety bill that would limit the federal government’s ability to halt the release of potentially dangerous AI systems. The company’s draft proposal, revealed by Politico, would also override some state-level AI safety laws—potentially leaving everyday people with fewer protections against the risks of advanced AI. If passed, this bill could affect how much say workers, parents, and local communities have over the AI systems shaping their lives.

What Happened

On October 16, 2026, Politico reported that Anthropic—a major player in the AI industry, valued at over $18 billion as of 2025—has been quietly pitching its own version of an AI safety bill to federal lawmakers. The draft legislation would limit the circumstances under which the federal government could block the release of new, advanced AI models. It would also preempt, or override, some state-level AI safety regulations, meaning local governments would lose the power to enforce stricter rules than those set at the federal level.

This move comes as tech companies race to shape how AI is regulated in the United States. Unlike the European Union, which passed the EU AI Act in 2024 to set tough standards for AI safety and transparency, the U.S. has yet to pass comprehensive federal AI legislation. In the absence of federal rules, states like California and New York have started crafting their own laws to protect residents from AI-related harms—ranging from job displacement to privacy violations and algorithmic bias.

Anthropic’s proposal would shift the balance of power away from both the federal government and the states, giving private companies more control over when and how new AI systems are released to the public.

Inside the Anthropic AI Safety Bill Draft: Thresholds, Blocking Orders, and Preemption

Anthropic's draft would set up a federal framework covering frontier artificial intelligence models trained above 10^25 computational operations, while preempting state disclosure, evaluation, and incident-reporting rules. Under the draft text, only a federal court can halt a release, and only after Commerce proves an imminent risk of more than 50 deaths or $1 billion in damage. It also mandates regular third-party evaluations for large developers, though critics argue the enforcement powers and independence safeguards are too weak.

The rules apply to any frontier model trained above 10^25 operations. A "large frontier developer" is one with more than $500 million in AI revenue or more than $1 billion in AI development spending. Developers meeting those definitions must engage an independent evaluator every 180 days, who must notify Commerce within 72 hours of conditions that could produce catastrophic harm. Evaluators conducting these assessments are paid by model developers, provided payment is not tied to results.

Enforcement centers on a blocking order that Commerce can request only through the U.S. District Court for the District of Columbia. Each resulting court order lasts 45 days. The judge must also find that no less restrictive measure exists, and the court cannot order a developer to disclose or delete model weights.

The draft would override state rules in one area. It bars states from enforcing rules that require covered developers to submit model disclosures, undergo third-party evaluations, or report safety incidents. The proposal preserves state tort law, child safety, deployment and use rules, whistleblower supplements, and procurement. Preemption falls away if the federal requirements are struck down.

Critics include Alex Bores, author of the RAISE Act, who said the draft falls short on blocking authority and evaluator independence, and Nathan Calvin, who said passing it would leave the public less protected than current law. Jim Love questioned how independent evaluators can be when the company they assess pays them. In response, Anthropic's Anthony Cimino said the preemption language would be revised to give states a role in evaluator standards and shutdown capacity, and that the thresholds were worth re-evaluating. The draft is dated September 29, 2026 and remains an industry proposal with no congressional sponsor, rather than a bill moving through Congress.

Source: aipolicy.tech's published text and summary of the draft.

How This Affects Everyday People

For most Americans, the details of AI regulation might seem distant or technical. But the stakes are real: these laws determine who gets to decide whether an AI tool is safe enough for your child’s classroom, your workplace, or your local hospital. If Anthropic’s bill becomes law, it could make it harder for government agencies to step in if an AI system is found to be unsafe or discriminatory. It could also prevent your state from setting its own higher standards for things like data privacy, algorithmic fairness, or job protections.

Imagine a school district in Massachusetts wanting to ban a new AI-powered grading tool because it’s shown to misgrade students from certain backgrounds. Or a hospital in Texas wishing to pause the rollout of an AI diagnostic system that’s making errors with patients of color. Under Anthropic’s proposal, local and state authorities might be powerless to act—even if the risks are clear—because the company’s bill would set a single, federal standard that’s harder to challenge.

For workers, this could mean fewer safeguards against AI-driven layoffs or surveillance. For parents, it could mean less control over how AI is used in schools. For students, it could mean being subject to untested or biased algorithms in education or testing. As AI systems become more common in hiring, healthcare, policing, and social services, the ability to set local protections matters more than ever.

The Bigger Picture

Anthropic’s lobbying push is part of a larger trend: big tech companies trying to write the rules that will govern their own products. In recent years, Microsoft, Google, and OpenAI have all spent millions lobbying Congress and state legislatures. According to OpenSecrets, tech industry lobbying on AI and related issues topped $130 million in 2025 alone.

This isn’t just about one company. The debate over AI regulation is heating up worldwide. The EU’s AI Act, for example, gives regulators the power to ban AI systems that pose “unacceptable risks” to public safety or fundamental rights. In contrast, Anthropic’s draft bill would make it harder for U.S. agencies to take similar action. This could leave Americans with weaker protections than people in Europe—or even in some U.S. states that have tried to get ahead of the issue.

The risk, critics say, is that companies like Anthropic are prioritizing their own interests over public safety. By limiting government oversight and overriding state laws, the bill could make it easier for risky AI systems to reach the market—and harder for regular people to push back. This is fueling a growing backlash against AI among workers, parents, and communities who feel left out of the decision-making process.

What You Can Do

The Bottom Line

Anthropic’s proposed AI safety bill could make it harder for governments—especially at the state and local level—to protect people from the risks of advanced AI. As tech companies push to shape the rules in their favor, it’s more important than ever for everyday people to pay attention, speak up, and demand a say in how AI is used in their schools, workplaces, and neighborhoods. The future of AI safety shouldn’t be left to the companies building these systems—it should be shaped by the communities they affect.

Primary source: Politico — referenced for fact-checking; this analysis is independent commentary by Jonathan West.
Found this useful?

More on this topic