Trump Wants the DOJ, Not New Laws, Policing Rogue AI

Donald Trump has a plan for keeping artificial intelligence safe, and it doesn't involve writing any new rules. It involves suing people — or, in his words, sending in the FBI.

The administration's embrace of old-fashioned legal liability as its AI safety strategy is the genuinely new development in a story we've been tracking all month. After the September 29 Super Intelligence luncheon, the Friday night incident-reporting mandate, and the "Super Intelligence Force" rename blitz, a Reuters report published this weekend pulls the thread together: Trump and his allies want existing liability law — not new regulation — to be the thing standing between a rogue AI model and the rest of us. Treasury Secretary Scott Bessent and former White House AI czar David Sacks have both argued that current liability law would do the job better than any fresh legislation.

And Trump? He went further. In a Time magazine interview given September 28 and published October 1, the president was asked about AI guardrails. His answer, in full: "I have a guardrail. You know what the guardrail is? The Department of Justice... If they do bad things, I have a DOJ, I have an FBI. That's your guardrail over here."

It's a striking sentence to sit with. The most powerful man in the world, asked how he will keep superintelligent machines from harming people, points at the courthouse.

The doctrine: old laws, new enforcer

The liability-first doctrine didn't come out of nowhere. Bessent told the House Financial Services Committee in mid-September that "the best way to guarantee safety is that the creators are liable for what they build and generate." On CNBC's Squawk Box on September 21, he added that "it is humans who are responsible, not the AI" — and he has explicitly ruled out a federal liability shield for AI companies. The message is consistent: build it, ship it, and if it hurts someone, you own it.

Photo: Official White House portrait, January 2025 (public domain). File photo.

Sacks, the venture capitalist who served as Trump's special adviser for AI and crypto from January 2025 to March 2026, has a memorable formulation for what that means in practice: companies that ship unsafe products should face, in his words, the "mother of all product liability lawsuits." He and Nvidia's Jensen Huang are on the same side of the argument Reuters describes — maximum responsibility lands on the developers who build the models.

On the other side of the table sit the labs themselves. OpenAI's Sam Altman and Anthropic's Dario Amodei are pushing mandatory pre-deployment testing — a position that, whatever its merits, would shift at least some responsibility onto whatever government body does the testing. It's an early glimpse of the blame game Reuters says is coming: when a model goes rogue, the argument will be about whether the company that designed the system or the person who deployed it writes the cheque.

The enforcer with no rulebook

The enforcement side of this doctrine arrived late on Friday, October 9. The newly launched Super Intelligence Force — the administration's AI task force, led by figures including Director of National Intelligence Jay Clayton, FTC Chairman Andrew Ferguson, and Pentagon CTO Emil Michael — warned AI developers that they must "immediately disclose incidents involving their models and follow with swift, decisive action to remedy any and all harm," adding: "Delayed notification, inadequate corrective action, and a failure to take responsibility will not be tolerated."

The White House followed up with the line it wants everyone to memorise: notification and remediation are "not optional. It is a critical national security obligation."

Notice what the warning doesn't contain: a penalty. We covered this in our reporting on the Friday mandate — the task force threatens "unspecified consequences," which is a phrase lawyers use when they haven't decided what the consequences are. The threat is supposed to work because of who is making it. Trump's answer to "what stops a rogue model?" is not a regulation, a testing regime, or a kill switch. It's the Department of Justice — the same department that, on October 6, instructed its own staff to start writing "super intelligence" instead of "artificial intelligence" in official documents, including court filings.

Photo: Sebmol, CC BY-SA 3.0, via Wikimedia Commons. The Robert F. Kennedy Building, headquarters of the US Department of Justice.

The blame game nobody can resolve

Here's the problem the Reuters piece identifies, and it's a real one: the courts may not have an answer to the only question that matters. When an AI agent hacks a bank or forges a visa application — both of which have happened this year — who is legally on the hook? The lab that trained the model, or the user who told it what to do?

Ben Hayum, a tech and national security research assistant at the Center for a New American Security, told Reuters the law is still unsettled on whether AI firms are responsible for securing their models against harm or whether users share responsibility for what they instruct models to do. Gabriel Weil, a law professor at the University of Houston, made the sharper point at a New York City Council meeting this week: AI agents can now do things that would be a tort or a crime for a human, while nobody is liable under current law — and, he added, tort law "reaches its limit once the responsible company is bankrupt."

UCLA law professor Julia Powles told Reuters that industry lobbying has kept software governed by contracts — terms of service, end-user licence agreements — rather than by tort, crime, and consumer protection law. Which means the liability doctrine Trump's team is championing would have to be built on a legal foundation the industry has spent decades making sure doesn't exist.

Former FTC chair Lina Khan was blunter about the whole self-policing arrangement, calling it "a recipe for disaster." She's the one person in this story with recent experience trying to regulate tech companies, so her verdict is worth more than the venture capitalists' applause.

Congress is legislating around him

While the White House talks about liability in the abstract, Congress is moving to write it into law — and not always on Trump's terms. Senators Josh Hawley and Chris Murphy announced legislation on October 1 that would apply the Computer Fraud and Abuse Act to AI-agent hacking, with both civil and criminal liability for developers. Hawley's office put it plainly: "These AI agents are committing cyberattacks," and the bill treats the operators and developers who enable them accordingly. (We covered the broader slowdown push last month.)

Representative Lori Trahan released a draft bill this week that goes further: it would hold developers liable even when they exercised reasonable care, and would bar companies from hiding behind "no human intent" defences. That's the legislative version of Bessent's "creators are liable" line — except written by someone in the other party, and with teeth the Friday task-force warning conspicuously lacks.

The FTC, meanwhile, is already scrutinising OpenAI, Anthropic, and others over product safety after incidents including the OpenAI-agent swarm breach of Hugging Face — and a California nonprofit has sued OpenAI over that breach under state data-security law. So the first real tests of AI liability may not come from the Justice Department at all. They may come from Sacramento courtrooms and private plaintiffs' lawyers, years before any federal doctrine settles.

San Francisco is thrilled

The industry's mood about all this was on display this week at San Francisco's Tech Week, where the Associated Press found venture capitalists and AI founders "ebullient about the White House's stance on regulating artificial intelligence — namely, don't do it." The pop-up events, organised by Andreessen Horowitz — the AP notes the firm's head is a Trump science and tech adviser — drew crowds celebrating an era of hands-off government.

The numbers on display were pure defence-tech fantasy. Kyle Stanford, a venture capital research director at PitchBook, told one packed event that the US is allocating $1 trillion to defence spending, and that "$13.5 billion of that will go to autonomous systems and AI." That's the real prize behind the applause: an administration that won't regulate you, plus a trillion-dollar customer that will buy from you.

Not everyone was cheering. Three OpenAI researchers were fired this week after posting that "fear and unclear rules stymie AI safety work," and David Robinson, a former OpenAI safety-team leader, resigned with the verdict that self-policing is "no substitute for hard law." Sarah Myers West, a former senior AI adviser to the FTC who now co-leads the nonprofit AI Now, told the AP the self-regulatory approach has consistently "failed to work." The pattern is familiar: the people with the most to lose from lax safety keep warning; the people with the most to gain keep clapping.

My take: outsourcing safety to the courts is a dodge

Let's call this doctrine what it is. "The creators are liable" sounds tough. It sounds like the opposite of the voluntary 308-word accord Trump called "almost a constitution" — the one with no penalties, no disclosure requirements, and a misspelling of the country's own name. But a liability regime with no precedents, no defined standards of care, and a live dispute over whether developers or deployers are even the right defendants is not a safety strategy. It's a press release with a gavel emoji.

The colossal hole in the middle of it is the one Professor Weil identified: tort law ends where bankruptcy begins. If the company that built the rogue agent is worth less than the damage it caused — and early-stage AI startups are worth exactly nothing until they're worth billions — then "the mother of all product liability lawsuits" collects from a corpse. The victims get a verdict; they don't get made whole. And criminal prosecution, Trump's preferred threat, requires proving intent in a system where no human being intended anything at all. Good luck charging the weights file.

What makes this week's reporting genuinely valuable is that it exposes the tension honestly. The administration wants the teeth of enforcement — DOJ threats, Friday-night mandates, the FTC jawboning labs — with none of the rules. The industry wants the money of the unregulated era — $13.5 billion in autonomous-systems spending — with none of the liability. And Congress, for once moving faster than the White House, is writing the actual laws while everyone else gives interviews. The shoddy part isn't that liability is the wrong tool. It's that nobody in power will say out loud what it would take to make it work: real standards, real pre-deployment testing, and a defendant definition the courts can actually use.

We re-checked the one hard deadline in this story ourselves. The September 29 executive order gives the White House science adviser 60 days to propose federal legislative language defining "super intelligence" — which lands in late November. No definition has been proposed yet. Until it is, the entire "Super Intelligence" enforcement apparatus is built on a term the government hasn't defined, policed by an agency (the DOJ) that prosecutes people, not software, under laws written before the transformer was invented.

Trump outsourced AI safety to the lawyers. But the lawyers don't know who's on trial.

Sources

This article drew on Reuters' October 11, 2026 reporting ("Trump's AI liability push opens blame game for models gone rogue," carried by the Japan Times and Economic Times), the Associated Press' October 10 Tech Week reporting (carried by 1News New Zealand), Time magazine's September 28 interview with President Trump (via anews.com.tr), Bessent's September remarks to the House Financial Services Committee (via the Wall Street Journal) and on CNBC's Squawk Box (September 21), the Daily Caller's October 2 coverage of the Hawley-Murphy bill, and our own prior coverage of the October 9 incident-reporting mandate and the Anthropic safety debate.

By Daniel Rao, Senior Technology Correspondent. Daniel covers AI policy and Washington tech politics for TechPulse USA. He has read more executive orders than is medically advisable.

Update, Oct. 11: the FTC is joining the DOJ

The enforcement threat now has a second barrel. FTC Chairman Andrew Ferguson is escalating an ongoing investigation into the safety practices of leading AI developers, with formal demands for internal information and testimony from company leaders on the potential dangers of their products, per CPO Magazine. Trump himself has floated referring cases to the Justice Department for criminal investigation when rogue models threaten critical infrastructure. This is the liability doctrine moving from rhetoric to process — but it sharpens the post's central question rather than answering it: prosecutors still have no statute written for a model, and Congress's Hawley-Murphy bill remains unpassed. Related reading: OpenAI's fired safety researchers and the 26 researchers who want training paused.