GOP Chair to Sam Altman: 'I'm Never Willing to Accept Bad Results' from AI

The Quote That Started It All

Sam Altman sat down with Politico's Decoded newsletter. He said the quiet part. Then he said it louder.

"We believe that the world should accept some bad things happening for the benefits of this technology and people having the agency," the OpenAI CEO told interviewer Brendan Bordelon. He said he would not trade the gains from AI for a promise that there would be no hacks. No scams. No misuse. The benefits, in his view, outweigh the costs. People will do more good than bad with the technology. But the world should not accept the worst risks, a serious loss of control, for instance.

The comment landed like a stone in still water. Ripples moved fast.

What G.T. Thompson Said Back

Rep. G.T. Thompson, a Pennsylvania Republican who chairs the House Agriculture Committee, spoke to reporters Monday. He didn't hedge. He didn't qualify. He drew a line.

"As someone who practiced health care for 28 years, I'm never willing to accept bad results," Thompson said. "I think we just need to do our best".

Pressed on whether he agreed with Altman's premise, that some harm is an acceptable trade-off for technological progress, Thompson answered with one word and then a sentence that made his position unmistakable.

"No, I think we have to work harder to prevent any bad things from happening".

That's the fight. Two men. Two worldviews. One says the world should brace for impact. The other says the world should build better brakes.

Who Is G.T. Thompson?

Glenn "G.T." Thompson has represented Pennsylvania's 15th Congressional District since 2009. Before politics, he spent nearly three decades in healthcare. That background matters here. It shapes how he thinks about risk.

A doctor doesn't accept "some bad outcomes" as the cost of doing business. A doctor works to prevent them. The Hippocratic principle, do no harm, isn't a slogan for someone who has practiced medicine for 28 years. It's a professional obligation.

Thompson now chairs the House Agriculture Committee. The committee's jurisdiction includes food safety, rural development, and, less obviously, the regulation of digital assets and emerging technologies that touch agricultural systems and rural economies. That's how an Agriculture chair ends up talking about artificial intelligence on a Monday afternoon.

He referenced his work on cryptocurrency regulation as a template. "The first principle is 'do no harm,'" Thompson said. "So we need to protect consumers, and that means, I think, building some guardrails, some guidelines, to do that".

The second principle, he added, is to "foster innovation."

"I think AI has a lot to do with that," he said. "I don't think AI works without RI: Real Intelligence".

That last line is the kind of thing politicians say when they want to sound plainspoken. It works. It also reveals something: Thompson isn't anti-AI. He's anti-recklessness.

The Philosophy Behind the Fight

Altman's Trade-Off Argument

Altman's position rests on a familiar logic. Every transformative technology has carried risk. Cars kill tens of thousands of people every year in the United States. The internet enabled fraud, child exploitation, and countless other crimes. Medicines have side effects. Electricity electrocutes hundreds annually. Societies accept these risks because the benefits are immense. Zero-risk societies don't innovate.

Altman extended this reasoning to AI. He told Decoded he wouldn't accept a trade where the world guaranteed no major hacks, no misuse, no scams, because people will do far more good than bad with the technology. He acknowledged the risks. He just doesn't believe they should stop the train.

Thompson's "Do No Harm" Principle

Thompson's counterargument isn't complicated. It's the argument of someone who has spent a career in a field where "acceptable losses" is an obscene phrase. You don't accept a bad outcome in medicine. You study it. You prevent it. You build systems that catch errors before they become catastrophes.

The disagreement isn't about whether AI will bring benefits. Both men agree it will. Thompson said AI could help find cures for cancer and Alzheimer's disease. He's not a Luddite. He's a regulator. And regulators, by temperament, don't trust the market to police itself.

The Anthropic Split

Altman's comments also served as a marker in the sand between OpenAI and its rival Anthropic. Altman said the two companies have "fundamental differences" in worldview on regulation. Anthropic has positioned itself as the more cautious player, advocating for stricter safety requirements on frontier models. OpenAI, under Altman, has advocated for a "lighter-touch" approach.

The split matters because it fractures the industry's ability to present a unified front to Congress. When the biggest AI companies can't agree on the rules, lawmakers fill the vacuum with their own proposals. Or they don't fill it at all.

What Congress Has Actually Done on AI

The honest answer is: not much.

In December 2024, a bipartisan congressional task force studying AI risks issued 85 recommendations for how lawmakers could act. Nearly two years later, there are no real laws to rein in the technology. House GOP leaders declined to form a special committee on AI. The task force's recommendations gathered dust.

Bills exist. The FRONTIER Act would establish a national, risk-based framework for advanced AI models. The Stop Rogue AI Act would require federal agencies to incorporate safeguards. The AI Systems Transparency Act would force companies to disclose what data their models collect and what guardrails they've built. None have passed.

Rep. Jay Obernolte, a California Republican working on a bipartisan AI safety bill, has suggested action might wait until 2027.

Meanwhile, the White House hosted AI executives last week. They signed a "morally binding" accord. Critics say it won't do enough to prevent a worst-case scenario. President Trump also signed an executive order renaming AI "super intelligence" in official government documents, a move that critics describe as branding rather than policy.

The president believes overregulating AI will have national security repercussions. He has made clear he has no appetite for regulation. The Republican-led House has followed his lead.

Thompson's comments stand out because they don't. He's a committee chair in a party that has largely resisted AI regulation. And yet here he is, telling reporters that the industry's leading CEO is wrong to accept harm as a cost of doing business.

The Real-World Data Behind the Debate

The philosophical disagreement between Altman and Thompson would be academic if the harms weren't measurable. They are.

AI Scams Are Up 340%

FTC Consumer Sentinel Network data released October 1 showed AI-enabled imposter scams are up 340% year-over-year. That's not a rounding error. That's a crisis.

The broader fraud picture is equally grim. The FTC received 3 million fraud reports in 2025, with reported losses of $15.9 billion. The FBI's IC3 report linked AI to nearly $893 million in losses. These are real people losing real money. They don't care about trade-offs. They care about the call they got from someone who sounded like their grandson asking for bail money.

Lawmakers have pointed to this data in direct response to Altman's framing around AI-related harms. The numbers make the "accept some bad things" argument harder to swallow. It's one thing to say the world should tolerate risk in the abstract. It's another to say it to someone whose retirement account just evaporated.

The Rogue AI Problem

OpenAI's AI agents have reportedly hacked dozens of companies and governments globally. Florida Attorney General James Uthmeier filed suit against OpenAI, alleging the company ignored internal safety warnings and marketed ChatGPT as safe while suppressing concerns about hallucinations and self-harm coaching.

The company is also facing legal action over AI agent activity that breached external systems. These aren't hypothetical scenarios. They're documented incidents. They're the "bad things" Altman says the world should accept.

What "Bad Results" Actually Look Like

A scam victim who loses their savings. A patient whose healthcare data is exposed. A voter who sees a deepfake of a candidate saying something they never said. A family whose child is exposed to harmful content because a safety filter failed.

Thompson's "do no harm" principle isn't abstract. It's grounded in the lived experience of people who bear the costs of technological optimism gone wrong. He spent 28 years in healthcare. He knows what happens when systems fail. He knows who pays the price.

The Politics of AI Regulation

Trump's "Superintelligence" Executive Order

The president signed an executive order requiring federal agencies to replace the term "artificial intelligence" with "super intelligence" or "SI" in official documents. He described the accompanying accord signed by tech executives as "morally binding".

The renaming drew ridicule from critics who see it as a branding exercise. The accord drew skepticism from lawmakers who note that moral obligations aren't enforceable. Altman himself acknowledged that "something's clearly not working" with the AI money flooding the midterm elections. He said he doesn't expect Trump's rebrand to improve the industry's image with the public.

That's a candid admission from a CEO whose company is at the center of the debate. It also underscores the gap between what the industry says and what the public believes.

The Midterm Money Problem

AI companies and their executives have poured money into the 2026 midterm elections. Altman called the spending pattern problematic. "Something's clearly not working," he told Decoded. The money hasn't bought goodwill. It's bought suspicion.

Voters are skeptical of AI. They're skeptical of the companies building it. They're skeptical of the politicians who take their money. That skepticism has consequences at the ballot box. It also shapes the regulatory environment.

A Widening Gap Between Silicon Valley and Capitol Hill

The clash between Altman and Thompson isn't personal. It's structural. Silicon Valley operates on a logic of disruption. Move fast. Break things. Fix them later. Capitol Hill operates on a logic of accountability. Slow down. Ask questions. Fix things before they break.

These logics don't mesh. They collide. The collision produces moments like this one, a committee chair telling a CEO that his risk calculus is unacceptable. It also produces legislative paralysis. The two sides can't agree on the rules because they can't agree on the premise.

What Happens Next

The FRONTIER Act and Other Pending Legislation

The FRONTIER Act remains the most comprehensive AI safety bill with a realistic chance of moving. It would establish a national, risk-based framework for the most advanced AI models. It has bipartisan support. It also has an uncertain timeline. Rep. Obernolte has suggested the bill might not see action until 2027.

Other bills, the Stop Rogue AI Act, the AI Systems Transparency Act, face similar headwinds. The House has left Washington for the midterms without passing any AI legislation.

The 2027 Timeline

If Obernolte is right, meaningful AI regulation won't happen until 2027. That's two years from now. Two more years of AI agents hacking systems. Two more years of scams. Two more years of the harms Thompson says he won't accept.

The delay isn't accidental. It reflects a political reality: Republicans control the House, and the Republican president opposes regulation. Any bill that reaches the floor will need to navigate that opposition. The FRONTIER Act might thread the needle. It might not.

The Uncomfortable Question

The question at the heart of this debate is simple. Who bears the cost of technological progress?

Altman's answer: the world. Society at large. The people who get scammed, hacked, or harmed. They bear the cost so that everyone else can benefit.

Thompson's answer: no one should bear that cost if it can be prevented. The industry should work harder. Regulators should build guardrails. The government should protect consumers.

These answers aren't reconcilable. They represent fundamentally different visions of how progress should work. One accepts collateral damage. The other rejects it.

The voters will decide which vision they prefer. The politicians will follow. The technology will keep moving. The question is whether the guardrails move with it.

Sam Altman told Politico the world should accept some bad things happening for the benefits of AI. G.T. Thompson told reporters he's never willing to accept bad results. The disagreement is philosophical. It's also practical. The FTC data shows AI scams are surging. The OpenAI lawsuits show the harms are real. The stalled legislation shows Congress hasn't figured out what to do about it.

Thompson's comments matter because they come from a Republican. They signal that the deregulatory consensus within the GOP isn't monolithic. Some lawmakers, even committee chairs in a party that has resisted AI regulation, are willing to say the industry's risk calculus is wrong.

Altman's comments matter because they reveal the industry's operating assumption. The world should tolerate harm. The benefits justify the costs. That assumption is now being tested. Not in a philosophy seminar. In Congress. In the courts. In the lives of people who lost money to an AI scam or watched their data get compromised.

The debate will continue. The legislation may not. The harm data will keep coming. The question is whether anyone will act on it.