The U.S. government halts Anthropic's most powerful model, causing the IPO stock price to plummet 3.7%

Author: Claude, Deep Tide TechFlow

Deep Tide Guide: The U.S. government, citing national security concerns, suddenly ordered Anthropic to shut down its two most powerful AI models, Fable 5 and Mythos 5, on June 12. In response, Hyperliquid perpetual contracts tracking its IPO expected valuation dropped as much as 3.7% on Saturday to around $1,627, a clear retreat from the over $1,800 high after the models' release. Anthropic publicly stated strong opposition, claiming the government only issued a verbal recall based on a "narrow jailbreak vulnerability," and if this standard is applied industry-wide, it would "basically halt all deployment of cutting-edge models."

A company founded on "AI safety" is being forced to shut down its most powerful products due to government security directives.

According to reports from CoinDesk and other media outlets, on the afternoon of June 12 (Friday), the U.S. Department of Commerce issued export control instructions to Anthropic, requiring it to suspend all access—regardless of whether users are U.S. citizens or foreign nationals, inside or outside the U.S., including the company's own foreign employees—to Claude Fable 5 and Claude Mythos 5. Unable to selectively enforce without affecting compliance, Anthropic chose to shut down these models for all global users, while other Claude models remain unaffected. Anthropic explicitly stated in a release that it does not agree with the government’s basis, calling it "a misunderstanding," and is working to restore access as soon as possible.

Directive targets national security, a letter triggers comprehensive shutdown

According to the Wall Street Journal (cited by Quartz), U.S. Commerce Secretary Howard Lutnick sent a letter to Anthropic CEO Dario Amodei listing the relevant restrictions. Anthropic revealed that it received the directive at 5:21 p.m. (Eastern Time) that day, but the letter did not provide specific details about national security concerns.

Anthropic said that, as understood, the government believes it has found a way to bypass—i.e., "jailbreak"—Fable 5. Company executives explained that the referenced vulnerability involves a narrow prompt engineering technique used to reveal existing, minor software flaws in the model’s code, and emphasized that similar outputs can be easily obtained from competitors like OpenAI without any special circumvention.

The language in Anthropic’s statement was unusually direct: "We are complying with the government’s legal orders and removing access to Fable 5 and Mythos 5 for all users. But we do not agree that discovering a narrow potential jailbreak should justify recalling a commercial model already deployed to hundreds of millions." The company warned that if this standard is extended industry-wide, "it would essentially halt all new model deployments by leading AI providers."

The company also pointed out that, so far, the government has only provided verbal evidence, with no written documents, technical reports, or formal disclosures through existing channels, and promised to publish further technical analysis within 24 hours to clarify the limited nature of the issue.

Immediate market reaction, Hyperliquid perpetual contracts plunge

The reason this regulatory news became a financial event is that Anthropic has not gone public, yet the crypto market has already created a "shadow pricing" space for it.

According to CoinDesk, on Saturday, Anthropic’s perpetual contracts on Hyperliquid fell about 3.7% to roughly $1,627, after having traded near the $1,800 high in the days following Fable 5’s release. The open interest in these contracts is about $8.6 million, relatively small compared to larger private market proxy tools, but still a noteworthy signal for a company that has not yet filed for an IPO.

This contract’s price roughly corresponds to an implied valuation of trillions of dollars for Anthropic—$1,638 per contract approximates an implied valuation of $1.638 trillion. Within 24 hours of the news, the contract’s 24-hour trading volume was about $190k, with a funding rate of 0.0056%.

The impact is not limited to the derivatives market. According to Bitcoin.com, the ANTHROPIC token issued on Solana by Prestocks, claiming to offer spot-like exposure via a special purpose vehicle (SPV) structure, dropped 9.11% in 24 hours to $655.32, with total liquidity pools shrinking 18.09% to about $126k. It’s worth noting that Anthropic had previously warned that any unlicensed tokenization or derivatives claiming to offer its equity exposure could be considered invalid, and these products are not endorsed or affiliated with Anthropic.

Fable 5 was a milestone just days before shutdown

The two models that were halted had only just been released days earlier.

According to multiple media reports, Fable 5 and Mythos 5 were launched around June 9, representing significant capability leaps in software engineering, scientific reasoning, visual tasks, and long-context processing. Fable 5 was Anthropic’s first public release of a Mythos-level model—positioned above their previous flagship Opus. Both models are built on the same underlying architecture, with the main difference being output control: Fable 5 has an integrated classifier to intercept responses in high-risk areas like cybersecurity, while Mythos 5 is open to a separately vetted trusted group.

According to MarkTechPost, Fable 5’s classifier is an independent AI system designed to detect potential misuse; when queries involve cybersecurity, biochemistry, or model distillation, it falls back to Claude Opus 4.8, with users being informed of the fallback, which occurs less than 5% of the time. Notably, Mythos 5 previously supported critical cybersecurity defenses through projects like Project Glasswing—raising questions about whether the temporary disablement might slow down infrastructure defenses.

Backlash against security narrative, new variables in IPO process

For Anthropic, the awkwardness lies in the timing and logic of this event.

TechCrunch headlined "Anthropic’s security warning may have just backfired," reporting that this company, which has built its identity on "responsible AI development," is now being asked to shut down its most powerful models based solely on verbal reports—a bitter pill, and the company openly expressed dissatisfaction.

This controversy comes at a sensitive stage in Anthropic’s IPO preparations. The company has recently secretly filed registration documents, positioning itself among leading AI players planning to go public. Although liquidity in tools like Hyperliquid is thin, the decline in contracts reflects investor caution regarding its recent valuation and IPO prospects. Such sudden regulatory actions add new uncertainties to an already volatile IPO process.

Anthropic stated it is actively communicating to resolve the issue and restore services quickly, viewing the directive as "a misunderstanding." The final outcome of this incident could influence future interactions between regulators and innovators, and shape how exit rules and safety assessments are applied to evolving AI technologies.