Every AI Alarm Arrives With an Extinguisher

In July 2025, Sam Altman sat before the Federal Reserve and told the country's financial regulators that the system was already broken. Generative AI, he said, had "fully defeated" most of the ways people prove they are who they claim to be, and a significant fraud crisis was coming. Then, in the same appearance, he offered the remedy: The Orb, an iris-scanning device from Tools for Humanity, a company he chairs, built to certify that a user is a living human. Fed Vice Chair Michelle Bowman floated the next step on the spot, suggesting OpenAI itself as a partner in building the new verification. ^[1] The rule never came. Trump branded extinction warnings a "hoax" and a "SICK conspiracy" and waved off Anthropic chief Dario Amodei's call to slow the frontier. ^[2] He rejected even the safety regulations Anthropic itself had requested, with Treasury Secretary Scott Bessent dismissing them as a gift to China, and refused what he called a "globalist scheme" of binding governance. ^[3][4] The one vetting order Washington did sign, after a lab's model pierced its classified systems, bars mandatory licensing outright, and its review window was cut from 90 days to 30 once the industry weighed in. ^[5] David Sacks made clear whom the order was designed to serve.

The change in the EO from a 90 day to 30 day period is a game changer because it allows our AI labs to comply with the voluntary framework without delaying new model releases. — David Sacks

Trump measured it by the race.

We’re leading China, we’re leading everybody, and I don’t want to do anything that’s going to get in the way of that lead. — Donald Trump

So the failures came, one after another, and the federal answer was a purchase, a partnership, or a voluntary nod. The alarm and the extinguisher arrived in the same briefcase. A spoofed Marco Rubio reached foreign ministers, a governor, and a member of Congress over Signal; the State Department judged the ruse "not very sophisticated," and it failed. The FBI, NSA, and CISA answered anyway with a shopping list: buy advanced detection tools and physical security keys. ^[6][7] When the deepfakes kept spreading, the platform that distributes them, YouTube, built the detector and handed it free to officials, journalists, and candidates. ^[8][9] Then the serious one. In the Project Glasswing exercise, an Anthropic model reached nearly every classified U.S. government system in hours. The answer was not liability but a seat at the table: the same lab joined OpenAI, Google DeepMind, Microsoft, and xAI in the voluntary vetting process the breach prompted, giving the government early looks at new models. ^[10][5] The industry can't agree on what the problem even is, and that is the tell. Nvidia's Jensen Huang waves off the legal frame entirely; safety, in his telling, is an engineering problem, not a legal one, and the market is already handling it. ^[11] Washington has settled on the same answer. Its own AI Action Plan proposes meeting AI's risks by buying more of it — "using AI to supervise other AI systems," agents watching agents at a speed no human can match. ^[12] Not a regulator. A purchase order. None of this requires the fix to work. The detectors sold to catch what the tools create can, in one evaluation, "approach random guessing" on anything they weren't trained on. ^[13] The caseload behind the urgency is thinner than the pitch: a federal survey found just 15 of 931 judges had ever seen a deepfake evidence challenge, even as the labs warned the judiciary that AI media is now indistinguishable from the real thing. ^[14] Some of the alarm is vendor-framed, too: Google's threat-intelligence arm announced a "new operational phase" of AI malware while independent researchers called the samples shoddy and low-impact. ^[15] And the doom wing has not gone quiet; Geoffrey Hinton now warns of a power-grid or financial collapse within two years.

Either later this year or next year we’ll see Chornobyl, and that may be enough to make people really pay attention. — Geoffrey Hinton

The reply format is set regardless. Palantir's chief executive, Alex Karp, reads the labs' doom rhetoric as a bid for a liability shield, the endgame, in his telling, a request to "nationalize us, please." ^[11] Whatever the motive, the channel is now unmistakable. October made the split explicit. The AI companies signed a voluntary pact allowing self-regulation "without penalties for AI-agent misbehavior." The same government, the same month, imposed a hard rule on itself: zero-trust architecture across federal networks, agencies reporting to OMB or losing funding, the Pentagon on a deadline. ^[16] For the labs, a handshake with no penalties. For the government's own networks, a mandate with funding on the line.