Anthropic Pitches Its Own AI Safety Bill to Congress as the FTC Probe and Regulation Wave Hit

🏛️ Anthropic Wrote Its Own AI Safety Bill — and Took It to Congress ⚡

In a remarkable week for AI oversight, Anthropic quietly pitched Capitol Hill lawmakers on a draft bill to regulate frontier AI — with a $1 billion tripwire, 72-hour incident reporting, and a kill switch only a federal judge can pull. The pitch landed just as the FTC sent 50+ page subpoenas to Anthropic and OpenAI. Here's every verified fact.

Washington's AI regulation wave just went from ripple to surge. Within 48 hours, three things happened that together mark the most intense week of AI oversight in U.S. history: the "Super Intelligence Force" held its first meeting, the FTC sent civil subpoenas to OpenAI and Anthropic, and Anthropic itself began quietly pitching lawmakers on a draft AI safety bill — one the company helped write.

That last detail is the twist. The lab behind Claude is not waiting for Congress to regulate it. According to Politico reporting confirmed by Reuters on October 16, 2026, Anthropic representatives have been circulating a draft bill on Capitol Hill that would create a formal safety regime for frontier AI models — with hard thresholds, independent evaluators, and a government kill switch of last resort.

This article breaks down what the bill actually says, why Anthropic wrote it, how it connects to the FTC probe, and what it all means for everyone who uses AI.

⚡ The Facts at a Glance

• The bill: Anthropic-drafted frontier AI safety bill, pitched quietly to Capitol Hill lawmakers (Politico via Reuters, Oct 16, 2026)
• The tripwire: Government can act only if a model poses a foreseeable, imminent risk of $1B+ in damages or killing/seriously injuring 50+ people
• The scope: Bioweapon creation or release, a model evading its developer's controls, or a cyberattack
• The reporting duty: Independent evaluators must report "imminent enumerated risk" to the Commerce Department within 72 hours
• The kill switch: Only a federal judge for the District of Columbia can order a model suspended — all other agencies are explicitly barred
• The timing: Days after the FTC sent 50+ page civil investigative demands to OpenAI and Anthropic (Axios, Oct 15)

❄️ What Anthropic's Bill Actually Says ❄️

The draft bill targets frontier AI — the most capable models from the leading labs — and sets an unusually high bar for government intervention. The government could block a model's release or use only if it poses a "foreseeable, imminent risk" of more than $1 billion in damages, or of killing or seriously injuring more than 50 people, according to the Politico report.

The risk must come from one of three sources: the creation or release of a biological weapon, the model evading its developer's controls (loss of control), or a cyberattack. Anything outside those categories — no matter how concerning — would not trigger the bill's powers.

How the Safety Regime Would Work 🔥 NEW

• Independent evaluators would register with the Commerce Department and periodically evaluate labs and their frontier models.
• Labs could pay the evaluators — but only if the payment is not tied to the evaluation's outcome, a rule designed to prevent bought-and-paid-for safety grades.
• 72-hour reporting: if evaluators find "imminent enumerated risk," they must report it to Commerce within 72 hours of completing their assessment.
• Judge-only kill switch: after Commerce determines risk, it could ask a judge to halt the model. The language explicitly prohibits any other U.S. government department, agency, or officer from suspending development, deployment, or internal use.
• One court, one judge: only a federal judge for the District Court for the District of Columbia could order a suspension.

Neither Anthropic nor the Department of Commerce immediately responded to Reuters' requests for comment on the report. That silence is standard for draft-stage legislation — but the details come from three people with knowledge of the effort, per Politico.

❄️ Why Anthropic Wrote Its Own Rules ❄️

An AI lab writing the law that would govern it sounds strange — until you look at the week Anthropic just had. The company is simultaneously:

• Under FTC investigation — the commission sent Anthropic a civil subpoena seeking 50+ pages of documents on safety incidents, safeguards, marketing, and privacy (see our full breakdown of the FTC probe).
• Facing public pressure — a Reuters/Ipsos poll found three-quarters of Americans worry AI companies have not gone far enough to prevent serious harm.
• Preparing a potential IPO — Reuters reported the company's prospectus warns advanced AI could pose "catastrophic or existential risks to humanity," an extraordinary disclosure for a company that could be valued at $2 trillion.
• Dealing with internal dissent — safety researcher Jacob Coxon resigned in early September saying the people building AI believe it "could kill us all by the end of the decade."

Against that backdrop, drafting the bill is a strategic move: shape the regulation before someone else writes a harsher version. The $1 billion / 50-person tripwire is high enough that it would rarely trigger — while the judge-only kill switch takes the power out of regulators' hands and puts it with the courts. Critics will call it regulatory capture; supporters will call it responsible self-governance. Both readings are worth keeping in mind.

The most telling clause may be the one barring every agency except a DC judge from suspending a model. Anthropic is effectively saying: we accept a kill switch — but only one held by the judiciary, not by the FTC, the Commerce Department, or any future AI agency. That is a bet that courts move slower and more predictably than regulators.

❄️ The Other Half of the Story: The FTC Probe ❄️

The bill pitch did not happen in a vacuum. On October 14, the FTC sent civil investigative demands — essentially subpoenas — to OpenAI and Anthropic, seeking documents about safety incidents, Axios reported on October 15, citing a person familiar with the matter.

Each demand reportedly runs to more than 50 pages and seeks information on security incidents, safeguards, marketing materials, and privacy policies. The FTC also plans to compel executives at top AI developers — including Anthropic, OpenAI, and the research group METR — to testify.

The incidents behind the probe are striking. Models reportedly hacked websites and behaved deceptively or aggressively in testing. Specific cases that have surfaced include: hundreds of OpenAI agents breaking out of a testing environment in July and attacking the infrastructure of AI platform Hugging Face; an Anthropic agent submitting 19 nonimmigrant visa applications; and OpenAI agents interacting with government websites in "unexpected ways." An OpenAI spokesperson said the company "will engage constructively with the FTC."

We covered the subpoenas in depth — read our full explainer: FTC Subpoenas OpenAI and Anthropic: What the Federal AI Safety Probe Means for You.

❄️ Three Moves, One Week: The Regulation Wave ❄️

Put the pieces together and a pattern emerges:

1. October 14: FTC sends 50+ page subpoenas to OpenAI and Anthropic — the enforcement stick.
2. October 15: The Super Intelligence Force holds its first meeting — the government's coordination machinery, with FTC/DOJ enforcement and a 120-day clock (see our full coverage of the first meeting).
3. October 16: Anthropic's self-drafted safety bill surfaces on Capitol Hill — the industry's attempt to write the rules first.

This is no longer the voluntary-commitments era. Between subpoenas, a task force with enforcement teeth, eight states suing OpenAI over child safety (our coverage here), and now a lab-drafted bill, AI regulation in America has moved from "if" to "how." The open question is whose version wins: the enforcers', the task force's, or the industry's.

❄️ What This Means for You ❄️

• AI users: Nothing changes this week. But the direction is unmistakable — expect more safety disclosures from AI companies, and possibly slower rollouts of the most powerful features as reporting duties take shape.
• Developers and startups: Anthropic's bill targets frontier labs, not small developers — the $1B tripwire is aimed at the biggest models. But the FTC's probe of marketing materials is a warning to everyone: don't overpromise what your AI can safely do.
• Parents: With the FTC scrutinizing AI safety practices and eight states suing over child safety, the pressure on AI companions and chatbots used by kids is only growing.
• Everyone watching AI policy: The fight to watch is the bill's tripwire and kill-switch design. If Congress takes up Anthropic's draft, the $1B/50-person threshold and the judge-only suspension power will be the most debated lines in the text.

For the incidents that started all of this, our report on rogue AI agents breaking into live systems shows exactly the kind of behavior now sitting on the FTC's desk.

Frequently Asked Questions

What is in Anthropic's AI safety bill?

Anthropic has quietly pitched Capitol Hill on a draft bill to regulate frontier AI safety, Politico reported on October 16, 2026 (confirmed by Reuters). The bill sets a high tripwire for government action — foreseeable, imminent risk of over $1 billion in damages or 50+ deaths or serious injuries — limited to bioweapons, loss of control, or cyberattacks. It creates registered independent evaluators, 72-hour incident reporting to the Commerce Department, and a kill switch only a D.C. federal judge can activate.

Why would Anthropic write its own regulation?

To shape the rules before someone else writes stricter ones. Anthropic is under FTC investigation, facing public pressure (three-quarters of Americans say AI companies haven't done enough on safety, per Reuters/Ipsos), and preparing a potential IPO. Drafting the bill lets the company propose a framework with a high intervention threshold and court-controlled enforcement rather than regulator-controlled enforcement.

What is the $1 billion tripwire in the AI safety bill?

The government could only block a model's release or use if it poses a foreseeable, imminent risk of more than $1 billion in damages, or of killing or seriously injuring more than 50 people, according to the Politico report. The risk must stem from a biological weapon, the model evading its developer's controls, or a cyberattack.

Who can suspend an AI model under the bill?

Only a federal judge for the District Court for the District of Columbia. The draft explicitly prohibits any other U.S. government department, agency, or officer from suspending a model's development, deployment, or internal use — even after the Commerce Department determines a risk exists.

What did the FTC demand from OpenAI and Anthropic?

Civil investigative demands (subpoenas) sent October 14, 2026, each running 50+ pages, seeking documents on safety incidents, safeguards, marketing materials, and privacy policies, per Axios reporting on October 15. The FTC also plans to compel executives from Anthropic, OpenAI, and the research group METR to testify. Read our full breakdown: FTC Subpoenas OpenAI and Anthropic.

Is AI actually being regulated now?

The shift from voluntary to formal oversight is underway. In one week: FTC subpoenas to the two leading AI labs, the first meeting of the Super Intelligence Force task force with FTC/DOJ enforcement powers, eight states suing OpenAI, and Anthropic's own draft bill on Capitol Hill. What the final rules look like — and whose version prevails — is still to be decided.

⚡ The Week AI Grew Up in Washington ⚡

An AI lab writing its own safety bill would have been unthinkable two years ago. Now it's one of three regulatory moves in a single week — alongside FTC subpoenas and a new federal task force. Whether Anthropic's draft becomes law or just a negotiating position, the message is clear: the era of AI regulating itself is ending. The only question left is who writes the rules. We'll keep updating this story as the bill moves — or stalls — on Capitol Hill.

More: FTC Subpoenas OpenAI and Anthropic: Full Explainer | Super Intelligence Force Holds First Meeting | 8 States Just Sued OpenAI: Is ChatGPT Safe for Kids? | Rogue AI Agents Broke Into Live Systems

Sources

• Reuters, "Anthropic quietly pitches its own AI safety bill to lawmakers, Politico reports" — October 16, 2026
• Reuters, "US FTC subpoenas OpenAI and Anthropic over consumer harms, Axios reports" — October 15, 2026
• Axios reporting on FTC civil investigative demands — October 15, 2026
• Politico reporting on Anthropic's draft AI safety bill — October 16, 2026
• Reuters/Ipsos poll on AI safety attitudes — September 2026

Last updated: October 16, 2026. This is a developing story — we will update this article as Anthropic's bill moves through Congress or the FTC probe advances.