Musk and Zuckerberg Killed Trump’s AI Safety Order. In Three Phone Calls.

The most consequential AI policy decision of 2026 was not made in Congress. It was not made in Brussels. It was not made after public consultation, expert testimony or democratic deliberation.
It was made in three phone calls. On a Wednesday night. Between three men and a president.

What Happened — The Weekend That Changed AI Regulation

Over the weekend: Musk, Zuckerberg, and Sacks killed Trump's draft AI safety executive order in three Wednesday-night phone calls.

The draft executive order — details of which have not been fully disclosed — was understood to include some form of AI safety requirements for frontier model developers. The three calls, made by Elon Musk, Mark Zuckerberg and White House AI czar David Sacks to President Trump, were sufficient to kill it.

No public debate. No congressional vote. No regulatory process. Three billionaires with direct access to the President of the United States decided that AI safety regulation was not in their interest — and it was gone by the weekend.
The contrast with Europe could not be more stark. The EU AI Act took four years of legislative process, hundreds of expert consultations, multiple parliamentary votes and thousands of pages of technical documentation to produce. In Washington, the equivalent was killed in an evening.

Anthropic Closes $30 Billion — The Same Saturday

Anthropic closed a $30 billion-plus round the same Saturday.

The timing is extraordinary. The company that has most consistently argued for AI safety regulation — whose CEO Dario Amodei has testified before Congress on AI risks, whose safety research is cited by European regulators, and whose model Claude Mythos was withheld from public release because its cybersecurity capabilities were deemed too dangerous — raised $30 billion on the same weekend that AI safety regulation was killed by phone call.

Anthropic's $30 billion round is the largest private funding round in AI history. It values Anthropic at approximately $100 billion. The investors backing it — betting $30 billion that Anthropic's safety-first approach is commercially viable — made that bet on the same weekend that the political environment for AI safety in Washington became significantly more hostile.
That is either a statement of extraordinary confidence in Anthropic's model — or a bet that the regulatory pendulum will eventually swing back.

Microsoft Cancels Its Claude Code Pilot — Quietly

Microsoft quietly cancelled its internal Claude Code pilot after token billing ate the entire annual AI budget, redirecting developers to Copilot.

This is the most revealing enterprise AI story of the week — and the least covered. Microsoft — which has invested over $100 billion in OpenAI and built its entire developer strategy around Copilot — ran an internal pilot of Anthropic's Claude Code for its own developers. The pilot was cancelled not because Claude Code performed poorly, but because the token-based billing consumed the entire annual AI budget faster than anyone anticipated.

The implication is significant for every enterprise evaluating AI coding tools: compute-based pricing for agentic AI development tasks generates costs that flat-rate subscription models do not prepare organisations for. Microsoft discovered this internally — and redirected to Copilot, which it controls and can price more predictably.

Every enterprise running Claude Code, GitHub Copilot Workspace or any other agentic coding tool on consumption pricing should be monitoring their token spend weekly. The budget risk is not theoretical. Microsoft just confirmed it is real.

The Three Stories Together — What They Mean

Read separately, these are three interesting AI news items. Read together, they are a portrait of the AI power structure in 2026.
AI safety regulation in the United States is determined by the preferences of a small number of extremely wealthy individuals with direct presidential access. The company most committed to AI safety just raised $30 billion — suggesting markets believe safety and commercial success are compatible even if Washington does not. And the world's most valuable enterprise software company discovered that AI cost models built on token consumption are unpredictable enough to destroy annual budgets.
The common thread: the AI era is being shaped by decisions made by very few people, with very little democratic accountability, at extraordinary speed.

The European Perspective