28 September 2026
Ro Khanna's AI bill would ban self-improving AI and create a new federal watchdog
Rep. Ro Khanna plans a bill that would pause AI that rewrites itself until federal safeguards exist, license frontier models, and add criminal penalties for disabling safety controls.
Congress has passed very little to regulate artificial intelligence. Rep. Ro Khanna, a Democrat who represents part of Silicon Valley, wants to change that with what he calls the most comprehensive AI safety bill proposed so far. It is named the Human Control Over AI Act, and a summary was shared first with CNBC.
What happened
The headline idea is a ban on recursive AI: models that improve themselves, or change their own core goals, containment or off switches, without people in charge. That ban would last until federal rules exist and an agency signs off. Other parts of the bill:
- A new federal agency focused on frontier models from OpenAI, Anthropic, Google DeepMind and xAI, with a licensing system for training and releasing them.
- Independent auditors placed inside every frontier lab, reporting straight to that agency.
- Standards for test sandboxes, air gaps (keeping test systems off the internet), kill switches, and oversight of the advanced chips the labs use.
- Criminal penalties for AI company employees who disable safeguards or knowingly release an unapproved system, plus mandatory liability insurance before a model launches.
- A push for export controls and agreements to deter China and other rivals from building dangerous AI.
Khanna told CNBC the bill draws on AI safety nonprofits rather than the wishes of big AI companies. He said the safety community should be "driving this debate."
Why it matters
The bill arrives as AI safety worries grow louder, including from some of the executives building the technology. It joins other proposals, such as the bipartisan FRONTIER Act from Reps. Jay Obernolte and Lori Trahan, which would also put auditors in frontier labs and let the government shut down models with catastrophic potential.
For investors, the details matter because AI now underpins a big part of the economy. Licensing, auditors, insurance and chip controls would all add cost and time for the labs and could shape demand for the chips they buy. A bill being introduced is a long way from a law, though.
What to watch
- The full bill text once it is formally introduced.
- Whether other lawmakers sign on as co-sponsors.
- Timing: no House AI bills are expected to get a vote until after the midterm elections.
Based on reporting by CNBC: Khanna to introduce AI safety bill with ban on 'recursive' technology until safeguards exist. Facts, not advice.
More news
- Anthropic's IPO, in numbers: 825 million shares at $70 to $78, and a $1.4 trillion price tag
- Trump says the Nvidia scandal didn't endanger US security, and stands by Lutnick
- Nvidia's first earnings without Huang: what Kress will be asked on Tuesday
- Buffett gives away $5.5 billion of Berkshire stock in his Thanksgiving letter
- Banks are using AI to trim back offices. Proof that it adds revenue is harder to find
- IMF chief says the world economy leans too hard on AI as circular deals top $1 trillion