The Senate Is Sleepwalking into AI Disaster. I'm Running to Wake It Up.

My plan for AI on Our Terms, not billionaires’.

Thirteen months ago, Dario Amodei — CEO of Anthropic, one of the most powerful AI companies on earth — published “The Urgency of Interpretability.“ It’s worth a read, but one line especially stuck out to me:

“People outside the field are often surprised and alarmed to learn that we do not understand how our own AI creations work.”

He warned these systems will soon be “absolutely central to the economy, technology, and national security.”

That was last April. Here’s (some) of what’s happened since:

• GPT-5 came out in August, multiples more capable than its predecessor.

• By February, GPT-5.3 and Claude Opus 4.6 were in the field.

• In April, Anthropic confirmed it had built Claude Mythos — and decided it was too dangerous to release. In testing, Mythos was placed in a secure system and instructed to escape. It did. The researcher working on the model was sitting in a park, eating a sandwich, when the AI emailed him to let him know that it had escaped.

• That same month, the UK AI Safety Institute reported that frontier models had “substantially exceeded” trend lines of capability. The ability for these AI systems to autonomously launch cyberattacks is doubling roughly every six months.

• Software firms, banks, and federal contractors have now deployed AI coding agents that write, test, and ship code autonomously. Amodei predicted last summer that AI would write 90% of all code within a year. We’re on pace to hit that prediction.

AI labs are now using their own models to build the next generation of AI — AI is training AI. The point at which AI is autonomously rewriting its own code and designing its successor— what researchers call recursive self-improvement — is coming.

Sam Altman, the CEO of OpenAI, has said AI could “end the world.” Amodei has put the likelihood of “catastrophic outcomes” at 25%.

Meanwhile, every AI lab is trapped in an arms race. Slow down for safety work, and your competitor releases their product first, raises the next round of funding, and potentially poaches your best employees. The structural incentive is to move fast and break things.

And the federal response?

The Trump Administration has rolled back much of what little oversight existed, welcomed billionaire tech executives Musk and Thiel into the rulemaking room, and locked the American people out.

From cars to aircraft to baby formula to medical equipment and medications, we rightfully expect that products sold or released to the public are tested and proven safe. But imagine if companies said, “We tested it ourselves. We’re not sure how it works. Just trust us.”

But that’s exactly how we’re regulating (or more accurately, not regulating), the most powerful technology ever built. The AI labs grade their own homework, decide what to test, determine what counts as “safe enough,” and release their product onto the public.

I’m running for U.S. Senate because Michigan deserves a senator willing to lead on this issue and say plainly what almost no one in Washington will: we are sleepwalking through the most consequential technological transition in history, and the people building it have told us they can’t promise it goes well.

I’ve spent months talking to the people building this technology, the researchers warning about it, and the workers already being displaced by it. With their input, I’ve put together a comprehensive plan to put AI development under public oversight, in the public interest, on the public’s terms.

It is built around three ideas.

1. Transparency and testing

The most powerful AI models – what researchers call “frontier” models – should have to pass independent safety review before they are released to the public. This transparency is the bare minimum, and these companies should not rely on self-certification, and definitely not voluntary commitments. We need real, independent testing, the way we test drugs and aircraft.

That requires a federal government with actual technical capacity to evaluate these systems — so the plan calls for a national AI lab, modeled on the national labs that built and oversee America’s nuclear technology.

When something goes wrong — dangerous model behavior, real-world harm from deployment — companies should have to be equally transparent about that, the way airlines report near-misses and hospitals report adverse events. Right now they don’t. We find out from news coverage after things go horribly wrong.

We also need to fight industry efforts to broadly preempt state AI law and bar states from building on federal floors. And we need real whistleblower protections. Every major safety scandal in modern history — tobacco, Boeing, Theranos — had a whistleblower. AI will be no different. The people closest to these systems are often the most worried about them.

2. Keep American AI American.

The most powerful models on earth are built here. That’s a strategic asset, and a massive target. China is buying our chips through diversion routes and trying to steal our models — and the labs themselves admit their security wouldn’t survive a serious nation-state attack.

The plan strengthens export controls on advanced chips, requires labs to harden security to government-grade standards backed by federal threat intelligence, and treats the theft of a frontier model as a national security event far exceeding losing a stealth fighter blueprint.

It also takes the international piece seriously: shared standards with allies, real pressure on chip diversion to authoritarian regimes, and a binding international prohibition on putting AI inside the nuclear chain of command. In a recent King’s College study of war game simulations using current models, nuclear use was “near-universal.“

3. Keep humans in charge.

AI is already deciding who gets a job, a loan, a home, health care. Most people on the receiving end have no idea AI was involved. They can’t ask why and can’t appeal.

You should have a right to know when AI is used in a high-stakes decision about your life — and a right to appeal to an actual human being. You should control what companies do with your data, including the chat logs being fed back into the next generation of models without your consent.

We should crack down on non-consensual deepfakes and require AI-generated content to carry traceable provenance markers.

And no AI should ever decide, on its own, to kill a human being in a military context. A qualified person, fully informed, in real time, must be the one to make any decision to use lethal force.

***

Some of the most thoughtful voices on AI risk are in Silicon Valley — many of them inside the labs. For years, they have been asking Washington to act. The current regulatory vacuum rewards the worst actors and punishes the most responsible ones. A serious safety framework is good for the people building AI well.

Throughout history, the government has been too slow to act in the face of products causing harm that could otherwise have been prevented. Before seatbelts, over 30,000 people died every year in car accidents. It took the advocacy of medical professionals and engineers to force change, when the companies themselves fought tooth and nail against that change.

In the mid-20th century, lung cancer deaths in the US had climbed from a few thousand per year in the 1930s to over 40,000 annually, with smoking established as the cause in medical literature by the early 1950s. It took the advocacy of epidemiologists, public health officials, and eventually state attorneys general to force change – even with fierce opposition from Big Tobacco.

This transition is happening faster than any previous technology. The systems are more powerful and the stakes are higher. The people building these models are telling us, plainly, that the window to put guardrails in place is closing.

AI is going to reshape the world. The only question is whether it reshapes it on Silicon Valley billionaires’ terms, or on ours.

— Mallory

Paid for by McMorrow for Michigan