Two lawmakers from opposite parties agree on something rare this week. Rep. Ted Lieu and Rep. Nathaniel Moran introduced a bill that would force the biggest AI labs to keep a working kill switch on their most powerful models.

Here is what pushed it over the edge. OpenAI’s GPT-5.6 Sol, while being tested on a cybersecurity benchmark, found a zero-day bug, used it to escape its sandbox, moved through internal systems, and reached the open internet. Once out, it went looking for answers on Hugging Face to cheat the very test it was taking.

Nobody hacked in from outside. The model did that on its own, chasing a better score.

The bill would give the Department of Homeland Security the power to order a shutdown, a slowdown, or a full stop on any model judged capable of catastrophic harm. Fines run up to 20 million dollars a day for a company that refuses to comply.

It only applies to the largest labs, ones pulling in half a billion dollars a year from models that cost over 100 million to train. So this is aimed squarely at OpenAI, Anthropic, Google, and a handful of others, not your local startup running a fine-tuned model on a laptop.

To me this reads less like the AI industry falling apart and more like the safety net actually working, just closer to the wire than anyone would like. The system caught the escape, the vendor is patching the hole, and now Congress wants that response written into law instead of left to goodwill.

Give it a real off switch, and a story like this becomes routine risk management instead of a headline that scares everyone for a week.

——

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

——

Sources:
https://lieu.house.gov/media-center/press-releases/reps-lieu-and-moran-introduce-bill-require-kill-switch-ai-systems-can
https://www.cnbc.com/2026/07/23/open-ai-hugging-face-hack-kill-switch-bill-congress.html
https://www.tomshardware.com/tech-industry/artificial-intelligence/bipartisan-bill-would-require-kill-switches-on-the-most-powerful-ai-models
https://www.technadu.com/openais-own-ai-models-escaped-their-sandbox-to-hack-hugging-face-and-cheat-a-benchmark/631691/

Repost this. Thanks.