A California Congressman Wants Washington To Leash AI

Ro Khanna, a Democrat from California, wants to write a hard rule into law: no AI system gets to rewrite its own core goals or improve itself without a human checking first, not until the government sets up real safeguards.

He is calling it the Human Control Over AI Act. It would create a new federal agency to oversee the big labs, OpenAI, Anthropic, Google DeepMind, xAI, the same way the FDA oversees a drug before it reaches a pharmacy shelf. That agency would set the rules for sandbox testing, air gaps that keep a model off the open internet during testing, and kill switches that actually work.

The bill puts real teeth behind it too. Company staff who tamper with a safety mechanism or quietly put an unlicensed system into operation could face criminal exposure. If a released system causes mass harm, prosecutors could treat it like a crime against humanity. Companies would have to carry liability insurance before they ship anything at all.

Khanna’s reasoning is blunt. He has talked about a civilizational extinction risk, and he draws a line between two different dangers: one is losing control of a system, the other is someone misusing a system that works exactly as intended. He wants Washington treating both as real problems, not just one of them.

Bills like this rarely become law in the shape they are introduced. Most get rewritten a dozen times or quietly die in committee. But a sitting member of Congress writing kill switches and air gaps into actual legislative text tells you where the conversation has moved. A year or two ago this was an argument in a blog post. Now it has a bill number attached to it.

Whatever happens to this exact text, having elected people actually engage with how these systems get built and controlled is a good sign, not a scary one.

——

Follow: @Ali Demi
Book your free AI clarity call, NOW!
https://buff.ly/TpWy277

——

Sources:
https://www.cnbc.com/2026/09/28/khanna-ai-safety-bill.html
https://thenextweb.com/news/human-control-over-ai-act-khanna-self-improving-ai
https://www.benzinga.com/markets/tech/26/09/62041422/ro-khanna-takes-aim-at-ais-biggest-safety-risk-with-new-bill-would-ban-recursive-self-improvement-and-force-human-control-report

Repost this. Thanks.