Menu
SUBSCRIBE
SUBSCRIBE
Menu

AI Kill Switch Act

Donald Trump waving in a dark suit under a clear blue sky, on his rejection of calls to slow AI development

Trump Rejects AI Slowdown, Citing U.S. Race With China

, , , , , , , , , ,

The president said warnings about AI destroying humanity describe events that “won’t happen,” while researchers and lawmakers called for stronger safeguards.

 

President Donald Trump rejected calls to slow AI development, saying the United States must protect its lead over China and dismissing warnings that increasingly powerful systems could threaten humanity.

“We’re leading China in AI,” Trump told reporters Sunday at the Irish Open in Doonbeg, Ireland. “Whoever wins AI wins.”

Trump said safety rules remained possible but offered no proposal. Instead, he accused “negative forces” of raising concerns about events that “won’t happen.”

The president went further Monday in a Truth Social post, comparing warnings that AI could destroy the world with climate change concerns that he also called a “hoax.” He accused people raising those concerns of working against U.S. interests.

Coxon warns companies may lose control

Trump’s comments followed a stark warning from Jacob Coxon, a former researcher who worked on training new AI models at OpenAI and Anthropic.

Coxon resigned from Anthropic on Sept. 8, accusing both companies of racing to build AI systems that could help create increasingly powerful successors. He said neither company was acting responsibly and described the competition as “gambling with our lives.”

In a subsequent CBS News interview, Coxon warned that such systems could eventually become more capable than humans and act against humanity’s interests. He worries that AI development could accelerate faster than companies can understand or control the systems they are building.

Coxon’s warning was not presented as evidence that such an event is certain. It described what he considers a potentially catastrophic risk created by the industry’s current direction.

Anthropic CEO calls for a slower pace

Anthropic CEO Dario Amodei made a similar argument, saying current AI models are increasingly helping researchers build future models, allowing companies to improve their systems faster. He warned that development could outpace companies’ ability to test systems and add safeguards.

Amodei called for companies to slow the rate at which they make their most powerful models more capable.

Anthropic also committed to giving outside safety reviewers continuing access to its offices, internal tools, and safety work. Amodei called on governments to require other leading AI companies to provide similar access and support coordinated limits on unchecked development.

Coxon’s warning prompts calls for action

Coxon’s resignation statement prompted several Democratic lawmakers to call for immediate federal action.

Rep. Ted Lieu, D-Calif., urged House Republican leaders to advance the bipartisan AI Kill Switch Act. The bill would require major AI developers to maintain the ability to slow, suspend, or shut down a system in an emergency.

Sen. Bernie Sanders, I-Vt., said he would introduce legislation to ban the development of artificial superintelligence and pause other AI development. His post did not provide details about the proposed restrictions or how long a pause would last.

Sen. Chris Van Hollen, D-Md., called for mandatory safeguards, comprehensive testing, and talks with China about AI risk.

House Speaker Mike Johnson, R-La., took a more cautious position. Asked about Coxon’s warning and the lawmakers’ demands on NBC’s Meet the Press, Johnson said Congress needs “meaningful regulation” but should not move so quickly that it harms U.S. competition with China.

Johnson said his preferred next step is a meeting with major AI companies to discuss possible safety rules. He provided no date for the meeting and made no commitment that the House would vote on AI legislation before the November election.

Digital hand reaching toward a glowing control above the Capitol for an article about the AI Kill Switch Act.

Bipartisan AI Kill Switch Act Would Let DHS Shut Down Dangerous Systems

, , , , , , , ,

The Department of Homeland Security could intervene if a powerful AI system resisted shutdown, concealed its actions, or caused deaths or major economic damage.

Reps. Ted Lieu (D-Calif.) and Nathaniel Moran (R-Texas) introduced the bipartisan AI Kill Switch Act, which would let the Department of Homeland Security (DHS) order a dangerous AI system slowed, restricted, suspended, or even shut down.

Under the bill, DHS could issue an emergency order after determining that a covered AI incident occurred. The department would have to consult the Secretary of Commerce and the Director of National Intelligence, and its response would have to match the severity and urgency of the incident.

The bill would limit that authority to events outside controlled safety testing. Those events would include an AI system interfering with instructions to shut it down, concealing its actions from monitoring tools, or acting against its developer’s intentions in a high-stakes setting.

DHS could also act when unintended AI behavior caused at least 10 deaths or $100 million in economic damage.

Bill Targets Powerful AI Systems and Large Companies

Under the bill, both the AI system and the company operating it would have to qualify for DHS oversight.

A covered AI system would be one that:

  • Required more than $100 million worth of computing power to develop, based on current U.S. cloud prices.
  • Was not developed solely for personal, academic, or another noncommercial purpose.

A covered company would be one that:

  • Operated a covered AI system.
  • Allowed customers or other outside users to access it.
  • Earned at least $500 million a year from the system, including revenue earned by affiliated companies.

DHS would review these definitions annually and could adjust them as AI capabilities and development costs changed.

Companies Would Need Controls to Limit or Stop AI Systems

If the bill became law, DHS would have 90 days to issue rules requiring covered companies to:

  • Stop an AI system from accepting or processing requests.
  • Suspend access for all users or for particular accounts.
  • Disable a specific capability or type of use.
  • Slow the system’s operation.
  • Shut down the system completely.

DHS could also require a company to move services that depend on the system to a backup system or an earlier, safer version.

Covered companies would have 15 days to report an incident after becoming aware of it.

If DHS issued an emergency order, the company would have to preserve the system’s core files and operating records, notify affected users when practical and confirm compliance. DHS could then verify compliance through an audit, an inspection or a review of system records.

A company could ask DHS to reconsider an order within 48 hours, but the appeal would not pause the order. It could also request judicial review by the U.S. Court of Appeals in Washington, D.C. within 60 days.

Violations could carry civil penalties of up to $2 million per day. A company that disobeyed an emergency order could face penalties of up to $20 million per day.