Rep. Ro Khanna, Democrat of California, leaves the U.S. Capitol after the last vote of the week, Thursday, September 3, 2026.
Tom Williams | Cq-roll Call, Inc. | Getty Images
Silicon Valley Democratic Rep. Ro Khanna will introduce a bill to regulate artificial intelligence that includes strict liability standards and a ban on self-improving “recursive” AI until the government creates safeguards.
The bill comes as Washington pushes to regulate AI after top executives warned the technology could spiral out of their control. Congress has done virtually nothing to regulate this technology, which supports the economy but could cause catastrophic damage if it goes wrong.
“There is actually a risk of extinction of civilization,” Khanna said Monday in an interview with CNBC. “There is a risk of loss of control for security, and then there is a risk of misuse, and we need to take both seriously.”
Khanna’s bill, a summary of which was shared exclusively with CNBC, is dubbed the “Human Control of AI Act” and would ban models that self-improve or recursively modify their own fundamental goals, containment or shutdown controls until federal safeguards exist and an agency approves the activities.
The bill would also create a new federal agency that would focus on AI models from OpenAI, Anthropic, Google DeepMind and xAI, as companies find themselves in the spotlight amid growing security concerns.
The agency would be responsible for creating security regulations, a licensing system to approve training and deployment of models, audits of border AI models, and establishing security standards for continuous testing and effective human oversight.
Independent auditors should also be embedded in each border laboratory and report directly to the agency. The agency would also create standards covering sandbox testing environments; air gaps or isolation of test environments from the Internet; circuit breaker; and controls that prevent models from escaping lab settings and moving across the Internet. It would also be responsible for regulating advanced chips used by pioneering labs, including mechanisms to monitor or control their use.
The bill would also create new legal penalties for AI malfunctions.
This would criminalize “crimes against humanity” related to the deployment of AI models resulting in the destruction of civilian populations. The bill would also impose criminal penalties on employees of AI companies who disable safeguards, kill switches, logging or containment systems, or who knowingly deploy an unauthorized system.
AI companies would also be required to carry extensive liability insurance to market their models.
The bill also calls on the administration to enter into enforceable agreements and control targeted exports to deter China and other adversaries from developing dangerous AI systems.
Khanna said the bill represents “the most comprehensive AI safety legislation” that has been proposed. He said this is inspired by his conversations with AI security organizations such as the independent nonprofits Model Evaluation and Threat Research, Machine Intelligence Research Institute and Palisade Research, rather than requests from executives of large AI companies.
“I want this to be a model for giving voice to the AI security community,” Khanna said. “I think they have been unfairly dismissed as science fiction and not taken seriously enough… it is up to them, the AI security community, to lead this debate, not the leaders of the border labs.”
It now joins a number of other proposals on AI safety, including the FRONTIER Act, a bipartisan proposal from Reps. Jay Obernolte, R-Calif., and Lori Trahan, D-Mass., that would place independent auditors in border labs and allow the government to shut down models that could pose a catastrophic risk.
No House bill is expected to be voted on before the midterm elections, and the Senate is expected to leave Washington and not return until after this week’s elections.
Gn headline
