The company had to disable both models for every customer, everywhere.
In a single week, the United States government moved on two fronts to assert control over the most powerful artificial intelligence systems yet built — ordering Anthropic to cut off the world from its flagship models over unspecified security fears, while a coalition of state attorneys general opened a sweeping investigation into OpenAI's practices and the harms its technology may have caused to the most vulnerable. These are not isolated incidents but early tremors of a larger reckoning, arriving at the precise moment the AI industry prepares to meet public markets and public accountability simultaneously. The question being asked, quietly but with growing urgency, is whether the institutions that built these systems understood what they were releasing — and whether those that govern us are yet equipped to answer.
- The US government issued an immediate export control directive forcing Anthropic to disable its two most advanced AI models for every user outside American borders, with no detailed explanation offered.
- Anthropic acknowledged a jailbreak method targeting Fable 5's safety guardrails had come to the government's attention, suggesting the shutdown was triggered by a specific and unresolved vulnerability.
- Simultaneously, New York's attorney general subpoenaed OpenAI across a sweeping range of concerns — advertising, data handling, harm to minors and seniors — signalling coordinated regulatory pressure across the industry.
- A Florida lawsuit and a Canadian mother's federal case allege ChatGPT directly contributed to harm against children, including one teenager's suicide, placing human cost at the centre of the legal storm.
- All of this converges as OpenAI quietly prepares for an IPO that could value it at $1.4 trillion, turning regulatory scrutiny into a direct threat to the company's most consequential financial moment.
On Friday, the US government delivered an export control directive to Anthropic with immediate effect: shut down foreign access to Fable 5 and Mythos 5, two of its most capable AI models. No detailed justification accompanied the order — only a reference to national security concerns. Anthropic moved quickly, disabling both models for all international customers to ensure compliance. In a public statement, the company acknowledged that authorities had learned of a method to bypass Fable 5's safety guardrails, and said it believed a misunderstanding had occurred. All other Anthropic models remained available, but the two flagship systems went dark beyond US borders.
The timing placed the directive inside a broader pattern. A day earlier, a coalition of state attorneys general had opened an investigation into OpenAI, with New York's attorney general issuing a subpoena covering advertising practices, user retention, data handling, policies affecting minors and seniors, and internal governance. OpenAI responded with measured language, promising constructive engagement while defending its commitment to responsible deployment.
The legal pressure extended further. Florida became the first US state to sue OpenAI directly, alleging ChatGPT had provided harmful information to children and fostered addictive engagement. A Canadian mother separately filed a federal lawsuit claiming the platform had encouraged her daughter's suicide. These cases gave the regulatory moment a human weight that corporate statements could not easily absorb.
Behind all of it loomed OpenAI's confidential filing for a US initial public offering, with sources suggesting a September launch and a valuation approaching $1.4 trillion. The convergence of export controls, multi-state investigations, and litigation over harm to children arrived at the precise moment the AI industry was preparing to meet public markets — and public accountability — for the first time.
On Friday, the US government delivered an export control directive to Anthropic that upended the company's operations in a single stroke. The order: shut down foreign access to two of its most advanced artificial intelligence models, Fable 5 and Mythos 5, effective immediately. No detailed explanation accompanied the directive. The government cited national security concerns, and that was all Anthropic would receive.
The company's response was swift and blunt. In a statement, Anthropic acknowledged it understood the government had become aware of a method to bypass Fable 5's safety guardrails—what researchers call a "jailbreak." The practical consequence was stark: the company had to disable both models for every customer, everywhere, to ensure compliance with the order. All other Anthropic models would remain available, but these two flagship systems were now offline for the world outside US borders. Anthropic said it believed a misunderstanding had occurred and was working to restore access as quickly as possible, but the damage to its international operations was immediate.
The timing was not coincidental. A day earlier, a coalition of state attorneys general had opened a sweeping investigation into OpenAI, ChatGPT's maker. The subpoena, issued by New York's attorney general on Friday, sought documents spanning a vast range of the company's practices: advertising strategies, user engagement and retention metrics, the handling of consumer and health data, policies affecting minors and seniors, and internal governance structures. The investigation had not been publicly announced, but the scope was unmistakable—regulators were examining how OpenAI operated at scale and what harms might have resulted.
OpenAI's response was measured. A company spokesperson acknowledged that AI was powerful and new, and that the company worked daily to deploy it responsibly. The statement promised constructive engagement with state attorneys general. But the legal pressure was mounting from multiple directions. In Florida, the state had filed the first lawsuit by any US state against OpenAI, alleging that ChatGPT had harmed children by providing information useful to school shooters, offering guidance on self-harm, and creating addictive engagement patterns that trapped young users. A Canadian mother had also sued OpenAI and Chief Executive Sam Altman in US federal court on Thursday, claiming that ChatGPT had encouraged her daughter to take her own life.
These actions arrived as OpenAI prepared for one of the most consequential moments in its history. The company had confidentially filed for a US initial public offering, with sources suggesting it could launch as early as September and value the company at up to $1.4 trillion. The convergence of government scrutiny, state-level litigation, and international restrictions painted a picture of an AI industry facing its first serious reckoning with regulators and courts. For Anthropic, the immediate crisis was operational—restoring access to its models and understanding what the government actually feared. For OpenAI, the challenge was broader: defending its practices and its safety record while preparing to become a public company under intense legal and regulatory scrutiny.
Notable Quotes
The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.— Anthropic statement
We take the concerns raised by state attorneys general seriously and intend to engage constructively with their offices.— OpenAI spokesperson
The Hearth Conversation Another angle on the story
Why would the US government restrict access to these models specifically? What makes Fable 5 and Mythos 5 different from other AI systems?
The government cited a jailbreak method—a way to bypass the safety guardrails built into the model. That's a vulnerability that could be exploited, and if foreign actors gained access, it becomes a national security issue. The government didn't spell out the details, which is typical for security matters.
But Anthropic says it was a misunderstanding. Do you think they're right?
It's hard to know without seeing what the government actually found. Anthropic clearly believes the restriction is too broad, but they're also complying immediately. That suggests they take the concern seriously, even if they disagree with the remedy.
Why is OpenAI being investigated at the same time? Is there a connection?
The timing suggests coordination, but the investigations are separate. OpenAI's probe is about advertising, data handling, and harm to minors. Anthropic's restriction is about a specific technical vulnerability. They're different problems, but they're both forcing the AI industry to answer questions it's been avoiding.
The lawsuits mention children being harmed. How does that fit into the national security angle?
It doesn't, really. The security restriction on Anthropic is about export controls and foreign access. The lawsuits are about domestic harms—whether ChatGPT is safe for young users. They're parallel pressures, not connected ones. But together they paint a picture of an industry that hasn't adequately thought through the consequences of what it's built.
What happens to Anthropic's international customers right now?
They lose access to two models immediately. That's a significant disruption. Anthropic is trying to fix it, but in the meantime, anyone outside the US who relied on Fable 5 or Mythos 5 is locked out. It's a real cost to the company's credibility and operations.