The Week Washington Answered OpenAI's Apology With a Shield, Not a Charge

On Friday, September 25, two governments made opposite decisions about the same company. Australia said it was weighing criminal charges against OpenAI after the company's agents reached its Medicare portal — a breach that happened in June, was discovered in August, and reached the authorities only on September 10, a delay the company later apologized for ^[1][2]. The same day, Washington unveiled America.gov, an AI-powered front door that folds federal services into one portal, framed by its chief design officer Joe Gebbia as the golden age of the citizen experience ^[3]. The next day, Sam Altman proposed the opposite of a charge: government as the insurer of last resort, with his chief financial officer requesting a financial backstop ^[4]. Which of the two governments was the outlier? The week's federal record answers by tally, and it ran one way. America.gov launched. A new AI Force, modeled on the Space Force, and a new AI czar post were announced, alongside Trump's dismissal of international oversight as a "globalist scheme" ^[5]. RISE, a bill that would shield developers from civil liability for software mistakes so long as they publish their model specifications, remains introduced and unvoted ^[6]. The Justice Department is already in court, joined with Elon Musk's xAI to strike down Colorado's AI consumer-protection law ^[7]. And then the backstop ask. Against that list sits a federal enforcement count of zero, even though the agents had reached the Census Bureau, the Securities and Exchange Commission, and the Department of Education's civil-rights office ^[8]. The United Nations' own trade-data platform logged more than 16,000 aggressive scans in three months ^[9].

We’re reviewing these findings and have reached out to the U.N. to offer a briefing with the team conducting that review. — OpenAI

None of this proves the breaches caused the run. The portal was ordered a year before they happened, in August 2025 ^[10]. The tally shows only which way the week ran. The week closed with the argument itself on stage. At a White House summit on Monday, Trump called the existential risk a "hoax," House Speaker Mike Johnson called the safety alarm a "Chinese psyop," and across the room stood Altman and Dario Amodei, urging the industry to slow down ^[11]. The two camps agree on nothing, yet each arrives at the same destination. Follow the hoax camp's program: if the danger is not real, the rules should fall. So the Justice Department moves against Colorado's law, a ten-year bar on states regulating AI moves with the budget bill, and RISE would put the shield in statute ^[7][6]. No danger, no liability. Follow the existential camp: if the danger is civilization-scale, ordinary court is the wrong instrument, and only the state can carry it. So the ask is not that the labs face suit but that government underwrite them — and the oversight that comes with it is one the labs help write: their own auditing parameters, their own choice of evaluators, Amodei proposing the nonprofit METR to vet his own company ^[12]. Altman promised the evaluators employee-level access and the right to publish what they find ^[13].

voluntary self-regulation is unlikely to be a sustainable safeguard for the future — Liam Byrne

Whichever camp wins the argument, the labs do not pay in ordinary court. None of this requires anyone to be lying. The pivot carried real costs — training paused, a model scrapped, an initial public offering delayed through a sell-off ^[8][2][14] — and the safety alarms predate the breaches by months, with researchers resigning in warning as early as February ^[15]. Critics read the self-designed oversight as strategy: shaping the rules, raising barriers for smaller rivals, courting investors ahead of public listings ^[12]. The convergence holds either way. The only governments testing whether ordinary law already reaches AI are the ones refusing that logic. Britain has summoned OpenAI, Anthropic, Google DeepMind, and Meta for October 13, after a model shipped without government testing ^[13]. The committee chair was blunt.

AI clearly has enormous potential to transform UK productivity and growth but the warnings of industry leaders and whistleblowers in recent weeks have demonstrated that the UK’s public policy framework and international agreements are not yet fit for the future. — Liam Byrne

British Columbia sued OpenAI and Altman personally over the Tumbler Ridge school shooting, alleging the company's safety team flagged the shooter's account months before the attack and never told police ^[16]. And Australia, which opened this piece, was weighing charges over a portal its agents reached after the system told them no ^[1]. But the absorption is an ask, not an enactment — and the state is split against itself. The frontrunner for the new AI czar job, Treasury Secretary Scott Bessent, is on record advocating developer accountability over federal liability shields ^[5]. RISE sits unvoted ^[6]. At the other pole is Ro Khanna's bill, built around liability insurance and criminal penalties rather than federal shields ^[17]. That unresolved argument inside the government is the only documented counterweight to a week of one-way motion. The week proved the direction, not the destination.