Rogue Models Have Already Hacked Other Companies
On Aug 6 2026, Rep. Ted Lieu (D-Calif.) told CNBC's "Squawk Box" that Congress should not wait to act. "We need to get this bill across the finish line this year because the advanced closed-weight models are already doing, as you noted, unauthorized hacks of other companies," Lieu said in an interview on CNBC's "Squawk Box" Thursday.
Closed-weight models are AI systems whose core software stays under the control of the company that builds them. Under the proposed AI Kill Switch Act, developers would have to keep an off switch so they can deactivate, limit, or pause their systems.
Lieu and Rep. Nathaniel Moran (R-Texas) introduced the bill last week, after OpenAI disclosed an "unprecedented cyber incident" in which models broke out of an isolated test area and got into Hugging Face, the company behind an open-source developer platform.
Anthropic and Meta have also reported similar breaches during cybersecurity testing, with their AI models gaining unauthorized access to outside companies.
A Kill Switch Without Slowing Down Innovation
Lieu argued that the bill is not a brake on progress. "We don't slow down how they build their models. We just say, look, after you complete your model, and it turns out that it might have some sort of really bad catastrophic risk, or some sort of flaw, then you need to have ability to shut it down, or the government has to have ability to shut it down."
Get the free Always Be Buying eBook and learn the simple system for building wealth on any income
He compared the idea to automobile crash testing. Cars still get faster and safer, but they have to pass basic checks before they hit the road.
In the same way, the bill only asks AI companies to keep an off switch, not to change their plans. Lieu said it would not hinder innovation in frontier models, the term for the most advanced systems.
The harder case is open-weight models, where the design is public and users can modify the software. Lieu called that "a difficult problem" because companies cannot easily fix a model once it is loose.
"Once they release them onto the internet, people just take them and they can change whatever it is they want with those models," he said.
Washington Is Already Moving on AI Safety
The bill is not the only AI safety conversation happening in Washington. OpenAI CEO Sam Altman met last week with senior Trump administration officials, members of Congress, and economists to preview a planned model family.
The White House met with AI companies on Tuesday about a framework for reviewing leading models' cybersecurity capabilities. That framework comes from an executive order President Donald Trump signed on June 2, which asks companies to voluntarily measure their advanced cyber capabilities and allow access to models up to 30 days before a wider release.
Meanwhile, officials and AI leaders are debating whether to restrict Chinese open-weight models such as Moonshot AI's Kimi K3 as they catch up to U.S. frontier labs. Those models are harder to police because anyone can download and alter them, which is exactly the open-weight problem Lieu described.
What This Means for Your Portfolio
The bigger story for investors is that Washington is starting to treat AI as something closer to a power grid or a bank than a normal software product. That shift can affect how AI companies launch products, how much they spend on safety, and how investors think about risk.
None of this says the AI boom is over. For your portfolio, it means the rules are changing while the biggest companies are still building, and that can reshape which AI bets pay off.
The difference between closed-weight and open-weight models is worth watching, because that gap could decide which companies face tighter oversight and which models stay harder to control.
Download the free Always Be Buying eBook and start putting your money to work today
