Congress Moves to Put AI on a Leash

What if an AI could rewrite its own code without anyone watching? That's the scenario a new bill from Silicon Valley Congressman Ro Khanna aims to prevent. According to CNBC, the Democratic representative is preparing to introduce the Human Control AI Act, a sweeping proposal that would put strict guardrails on frontier AI models.

The core of the bill is simple: no AI model can perform recursive self-improvement or autonomously rewrite its core objectives, suppression methods, or shutdown controls until the government sets up a security mechanism and a regulatory body gives the green light. In other words, companies won't be allowed to let their most powerful models iterate on their own—not without federal approval.

A New Agency with Teeth

To enforce these rules, the bill would create a new federal agency dedicated to AI safety. This agency would run a licensing system for training and deploying models, conduct audits on frontier systems, and set safety standards to keep them under human control. Every frontier lab would also need independent auditors who report directly to the agency.

The agency would maintain a specific safety checklist: sandbox testing environments, physical isolation from the internet, emergency shutdown switches, and measures to prevent models from escaping the lab and running wild online. Even high-end chips in labs would fall under regulation, with mechanisms to monitor or control their use.

Criminal Liability for Bypassing Safeguards

The bill doesn't hold back on punishment. If a deployed AI causes destruction to civilians, it would be classified as a crime against humanity, with criminal liability attached. Employees who privately disable protective measures, emergency switches, logging systems, or containment systems—or who knowingly deploy systems without authorization—would also face criminal charges. Companies wanting to release models would first need comprehensive liability insurance.

Khanna calls this the most comprehensive AI safety legislation proposed so far. He says its approach came from discussions with independent non-profit AI safety organizations, not from big tech executives. "I hope it becomes a model, giving the AI safety community a voice," he said. "They have been unfairly regarded as talking about science fiction and not taken seriously. This discussion should be led by the safety community, not by executives of cutting-edge laboratories."

Key Points

No self-improvement without approval: AI models can't recursively improve or rewrite core controls until a government safety mechanism and regulator sign off.
New federal agency: Would license model training and deployment, audit frontier models, and set safety standards.
Independent auditors: Each frontier lab must have auditors reporting to the agency.
Safety checklist: Sandbox testing, physical isolation, emergency shutdowns, and chip controls.
Criminal penalties: Deploying a destructive AI could be a crime against humanity; employees who bypass safeguards face charges.
Liability insurance: Companies must hold comprehensive coverage before releasing models.

Congress Moves to Put AI on a Leash