Trump Kills AI Safety Review as Hyperscaler Debt Hits Trillions

Washington Quits Before the Fight Starts

The federal AI safety review is dead. Trump's new executive order reverses his earlier position and removes the review process industry lobbyists spent months complaining about. There was no dramatic battle, no compromise language, no fig leaf. The administration simply decided the friction wasn't worth it and handed the win to the labs.

This matters because the safety review was the last meaningful federal checkpoint before deployment of frontier systems. With it gone, the only remaining oversight lives in state legislatures, in the FTC's increasingly thin enforcement budget, and in whatever voluntary commitments the labs feel like honoring this quarter. The honor system, in other words.

The lobbying that produced this outcome is itself becoming a story. Super PACs funded by rival AI factions are pouring money into congressional primaries, with a notable share going to anti-immigrant advertising aimed at shaping which Republicans survive to the general election. The factions disagree on plenty, but they agree that the people writing AI policy in 2027 should be people they picked.

The Money Is Past the Point of Embarrassment

Hyperscalers are raising debt at a pace that has turned Wall Street's structured products desks into the happiest place on earth. Hundreds of billions in AI-linked bonds, loans, and derivatives are now sloshing through the system, and the banks underwriting it all are running victory laps disguised as earnings calls. Whether the underlying compute investments earn their cost of capital is a question for later. Much later.

Musk, Altman, and Amodei are now openly racing toward public listings. xAI, OpenAI, and Anthropic all want a piece of the deepest pool of capital on earth before the narrative shifts or a competitor gets there first. An IPO is, among other things, a way to convert a private valuation argument into a public one, which is helpful when your private investors are getting tired of marking up the same shares every six months.

The combined picture is straightforward. The labs need staggering amounts of money, the banks are delighted to provide it, the federal government has stopped asking inconvenient questions, and the political machinery is being quietly rewired to keep it that way. There is no brake pedal in this car. There may never have been one.

The Products Themselves Are Getting Weirder

Google's Gemini can now generate video deepfakes convincing enough that a journalist used it to recreate one of Google's own advertisements, casting a stuffed deer as the vacationing protagonist. The point of the exercise was that the tool's outputs are no longer distinguishable from the marketing material the company used to sell the tool. The ouroboros is fully operational.

Surgeons are reporting a new category of patient: people asking to be reshaped according to AI-generated images of themselves. The proportions are not human. The surgeons, to their credit, are concerned. The patients, to nobody's credit, are persistent. This is what happens when the mirror starts editing back and people decide the mirror is the authority.

An author writing a book about AI's effect on reality included fabricated quotes generated by a chatbot, then blamed the chatbot when caught. The book's thesis survives the incident, arguably strengthened. Its author does not.

The Labor Market Catches Up

Computer science graduates are now experiencing unemployment rates higher than several humanities majors, philosophy included. The discipline that was supposed to be future-proof is discovering that when you automate the entry-level rungs of a profession, the entry-level workers do not magically appear at the senior rungs. They appear at the unemployment office.

This is the first cohort to graduate into a market where the tool they spent four years learning to use is also the tool replacing the job they trained for. Career services offices are reportedly recommending pivots into prompt engineering, sales engineering, and, with a straight face, philosophy. The students are not amused, and the philosophy majors are not making room.

The through-line of the day is that every check on this technology, regulatory, financial, epistemic, professional, is being removed or overwhelmed at roughly the same speed. The deer is on vacation. The author is on tour. The grads are on LinkedIn. The safety review is in the trash.