The Day Anthropic's Crown Jewels Went Dark
The biggest story today requires a double-take: Anthropic, which just released what may be the most capable publicly available AI model in history, has been ordered by the US government to pull its two most powerful systems — Fable 5 and Mythos 5 — from global access. The directive, framed around export control and national security concerns, targeted foreign nationals. Rather than build a patchwork of country-by-country restrictions, Anthropic made the starker call: disable both models entirely for everyone outside the permitted perimeter.
The timing is almost cruelly ironic. Fable 5 had just posted an 88% score on FrontierMath, the research-level mathematics benchmark widely considered one of the hardest tests in AI evaluation — putting it 13 points ahead of GPT-5.5. It also topped the Artificial Analysis Intelligence Index at 64.9, built on Anthropic's new "Mythos" architecture. By any technical measure, this was Anthropic's most significant launch. It is now also its most restricted.
The episode illustrates a tension that's been building for months: frontier AI capability and US export policy are now on a direct collision course, and companies may not get to choose how they land.
OpenAI Under the Microscope
While Anthropic's week was dramatic in its own way, OpenAI's Friday ended with a subpoena. A coalition of 42 state attorneys general has opened a formal multistate investigation into the company, demanding documents related to its operations, user safety, and data practices. The probe — reported by the Wall Street Journal — arrives at a particularly fraught moment: OpenAI is navigating its ongoing corporate restructuring, facing scrutiny over its nonprofit-to-for-profit conversion, and now contending with coordinated legal pressure from nearly the entire country's state-level law enforcement apparatus.
The subpoena's breadth suggests this isn't a narrow compliance check. Safety concerns and data practices together point to a wide-ranging inquiry that could surface internal communications at exactly the moment OpenAI would prefer to be managing its narrative toward an eventual IPO.
Meta's 6,500-Person Problem
Inside Meta, a different kind of crisis is unfolding. The company assembled its Applied AI team in March — roughly 6,500 engineers tasked with powering the Superintelligence Labs push — and three months in, the unit is in open revolt. Reports from Wired and TechCrunch describe a workforce frustrated by what they're calling "soul-crushing" busywork: tasks that feel disconnected from meaningful AI development, more support-ticket work than frontier research. For a company that has staked considerable prestige on competing at the top of the AI stack, a morale collapse inside its largest technical unit is a material problem, not just a culture footnote.
The Cost Wall Is Sending Enterprises Elsewhere
Beyond the headline companies, a quieter structural shift is accelerating: enterprise AI buyers are defecting. Frontier model costs have grown steep enough that a meaningful share of businesses are now turning to Chinese large language models and open-source alternatives to manage their budgets. This isn't ideological — it's arithmetic. When the bill for running AI at scale becomes a line-item problem, procurement teams look for substitutes. The trend complicates the revenue picture for US frontier labs and adds a new dimension to the national security debate: cost pressure may be doing what adversaries couldn't.
$130 Billion in Data Centers, Stopped Cold
The physical infrastructure underpinning the AI boom is running into a wall of local resistance. According to reporting from Ars Technica, community protests have blocked approximately $130 billion in planned AI data center construction. The backlash is hyperlocal — neighbors, municipal councils, and environmental groups opposing power consumption, water use, and the sheer industrial footprint of hyperscale compute — but its aggregate effect is national in scale. The build-out that AI companies assumed would proceed on their timeline is being renegotiated block by block.
Hollywood Edges Forward, SpaceX Lights the IPO Path
On the margin, two smaller but notable developments. A film company is reportedly the first to produce a project meaningfully shaped by custom-trained AI models rather than generic chatbot prompting — a quiet signal that Hollywood's AI integration, long promised and long delayed, may finally be finding a workable form. And SpaceX's IPO debut is already being parsed by investors as a template for what an OpenAI or Anthropic public offering might look like: a high-valuation, mission-driven company with a long runway and a loyal retail following. Both AI giants are watching the playbook closely.
Today's throughline: American AI is simultaneously at its most capable and most constrained — technically ascendant, legally embattled, and increasingly hemmed in by the geopolitical decisions of the government that helped build it.