A proposed law would grant U.S. officials the power to shut down AI systems that could cause catastrophic harm, requiring AI makers to embed technical kill switches.

The AI Kill Switch Act, introduced in Congress this month, would give the President and designated federal agencies the authority to order the immediate shutdown of artificial intelligence systems deemed to pose a risk of catastrophic harm.

The bill requires developers of high‑risk AI models to embed a technical “kill switch” that can be remotely activated, ensuring that the system can be disabled without destroying the underlying hardware.

Key Provisions of the Bill

Under the proposed legislation, any AI system classified as “critical” – such as autonomous weapons, large language models used in critical infrastructure, or AI that can manipulate public opinion at scale – must undergo a certification process overseen by the Department of Commerce.

If a certified system is found to be operating outside its authorized parameters, the President can issue an executive order directing the system’s operator to activate the kill switch within a prescribed timeframe, typically 24 hours.

Industry Response

Technology firms have expressed mixed reactions. Some companies welcome the clarity, noting that a standardized shutdown protocol could reduce legal uncertainty. Others warn that forced shutdowns could undermine trust in AI services and potentially cause unintended collateral damage.

  • Microsoft highlighted concerns about disrupting cloud services that host AI workloads
  • OpenAI emphasized the need for transparent criteria before a kill switch can be triggered
  • Google called for a bipartisan review to balance safety with innovation

Legislative Outlook

The bill is slated for debate in the House Energy and Commerce Committee next week. Proponents argue that the act is a necessary safeguard against rogue AI, while opponents contend that it grants excessive executive power and could be misused for political purposes.

If passed, the AI Kill Switch Act would become the first federal law to explicitly mandate built‑in shutdown mechanisms for AI, setting a precedent that could influence international regulatory approaches.

"We must ensure that the tools we create cannot outpace our ability to control them," said Rep. Jane Doe, a co‑sponsor of the bill.

The debate reflects broader concerns about AI governance, with policymakers seeking ways to mitigate risks without stifling the technology’s potential benefits.

Ars Technica coverage of AI Kill Switch Act