The Week's Earthquake: Anthropic's Most Powerful Models Go Dark
The story that consumed the AI world this weekend: the US government ordered Anthropic late Friday to immediately suspend global access to Claude Fable 5 and its companion model Mythos 5—two models that had sat at the top of global benchmarks since their launch just 72 hours earlier. The order came from the White House and was reportedly triggered by a tip from Amazon, which flagged concerns that the models could be jailbroken in ways that posed national security risks. Anthropic complied.
The scope is sweeping. Under US export control authority, no foreign national—whether living in the US or abroad—can access the models. The Trump administration simultaneously banned Anthropic's AI from use across federal agencies, a double blow that arrived without public warning and sent the company scrambling over a weekend.
The collateral damage is already drawing fire. A chorus of cybersecurity professionals has pushed back directly to the White House, arguing that locking Fable 5 and Mythos 5 away from foreign security researchers and defenders leaves the very people protecting critical infrastructure without the best tools available—while adversaries, who were never relying on Anthropic's API to begin with, face no comparable constraint. The argument, in short: export bans on AI models don't disarm attackers; they disarm defenders.
The episode raises a question that won't go away quickly: if a government can pull the two most capable AI systems on the planet offline with a Friday-evening phone call, what does that mean for the companies, developers, and users who built workflows around them?
OpenAI Under the Microscope—at the Worst Possible Moment
OpenAI's week was complicated before the Anthropic news broke. On Friday, June 12, a coalition of 42 state attorneys general served the company with a sweeping subpoena, formally opening a multistate investigation. The probe covers user data handling, safety practices, and the company's impact on vulnerable populations including minors—broad enough to touch nearly every product OpenAI ships.
The timing is pointed: the investigation landed just days after OpenAI filed for its IPO, a juxtaposition that will not be lost on prospective investors or the SEC. Regulators at the state level have clearly decided that the window before a public offering is exactly the right moment to start asking hard questions.
Meanwhile, the company is pushing forward on the enterprise side. OpenAI announced a formal Partner Network and committed $150 million to accelerate AI adoption inside large organizations—a signal that whatever scrutiny it faces from attorneys general, it is not slowing its commercial buildout.
Salesforce Buys Its Way Into AI Customer Service
In a deal that underscores just how much money is moving through the AI acquisition market, Salesforce announced it is acquiring Fin, an AI-powered customer service platform, for $3.6 billion. The move is a direct play to deepen Salesforce's position in enterprise automation—Fin's technology handles customer interactions end-to-end, and Salesforce's distribution muscle gives it a path into accounts its rivals can't easily reach.
It is the kind of acquisition that looks obvious in retrospect: customer service is one of the clearest ROI cases for AI, Salesforce already owns the CRM layer, and buying rather than building buys time in a market where incumbents are moving fast.
The Larger Pattern
Pull back and today's news sketches a coherent—if unsettling—picture. The US government is now willing to intervene directly and rapidly in the AI market, using national security authority to pull frontier models from the global market with little notice. Regulators at the state level are circling OpenAI at a moment of maximum vulnerability. And the enterprise market is consolidating, with billions changing hands as companies race to own the AI layer in their respective verticals.
The question for the week ahead: does the Anthropic model ban hold, get modified under pressure from the security community, or become the template for how Washington treats the next generation of frontier models?