Rep. Ro Khanna to introduce AI safety bill banning 'recursive' models

Rep. Ro Khanna, D-Calif., will introduce a bill to regulate artificial intelligence, with strict liability standards and a ban on "recursive" self-improving AI until the government creates safeguards. A summary was shared exclusively with CNBC.

The proposal lands as Washington faces growing pressure to act, after senior technology executives warned that advanced AI could slip beyond human control. Congress has so far passed no comprehensive rules for the technology.

Guarding against loss of control

"There's actually a civilizational extinction risk," Khanna said in an interview with CNBC. "There's a safety risk of loss of control, and then there's a misuse risk, and we need to take both seriously." The bill, the Human Control Over AI Act, would ban models that recursively self-improve or autonomously modify their own core objectives, containment or shutdown controls until federal guardrails exist and an agency approves those activities.

A new agency for frontier AI

The legislation would create a federal agency for frontier models from OpenAI, Anthropic, Google DeepMind and xAI. It would write safety rules, license model training and deployment, audit frontier models and set standards for continuous testing and effective human control. Independent auditors would sit in every frontier lab and report directly to the agency. Standards would also cover sandbox testing, air gaps that isolate testing from the internet, kill switches and controls that keep models from escaping lab settings. The agency would also regulate frontier labs' advanced chips.

Penalties, insurance and politics

The bill would criminalize "crimes against humanity" for deployments that destroy civilian populations and impose criminal penalties on AI employees who disable safeguards, kill switches, logging or containment systems, or knowingly deploy an unauthorized system. Companies would have to carry extensive liability insurance before releasing models. It also calls for enforceable agreements and targeted export controls to deter China and other adversaries.

Khanna called it "the most comprehensive AI safety legislation" proposed so far, saying it was modeled on talks with the nonprofits Model Evaluation and Threat Research, the Machine Intelligence Research Institute and Palisade Research rather than the requests of large AI companies. "I want this to be a model to give voice to the AI safety community," he said. The bill joins other proposals, including the bipartisan FRONTIER Act from Reps. Jay Obernolte, R-Calif., and Lori Trahan, D-Mass. No House vote is expected before the midterm election, and the Senate is expected to leave Washington after this week.