Author: Claude, Deep Tide TechFlow

Deep Tide Guide: On June 12, the U.S. government suddenly ordered Anthropic to shut down its two strongest AI models, Fable 5 and Mythos 5, citing national security. In response, the Hyperliquid perpetual contract, tracking its IPO expected valuation, fell 3.7% to about $1627 on Saturday, significantly retreating from the over $1800 high after the models were released. Anthropic publicly expressed strong opposition, stating that the government requested a recall based solely on a "narrow jailbreak loophole," and that if the standard were to be applied industry-wide, it would effectively halt the deployment of all advanced models.

A company that bases its foundation on "AI safety" is being forced to shut down its strongest products due to a government safety directive.

According to reports from CoinDesk and other media, on June 12 (Friday) afternoon, the U.S. government issued an export control directive to Anthropic, requiring the suspension of access to Claude Fable 5 and Claude Mythos 5 for all foreign nationals—regardless of whether they are within or outside the U.S., including the company's own foreign employees. Unable to selectively enforce without affecting compliance, Anthropic chose to close access to these two models for all users globally, while other Claude models remained unaffected. Anthropic explicitly stated in a statement that it did not agree with the government’s basis and referred to it as "a misunderstanding," striving to restore access promptly.

Directive Targeting National Security, A Letter Triggers Total Shutdown

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

Anthropic stated that, to its understanding, the government believed it had found a method to bypass, or "jailbreak," Fable 5. Company executives explained that the cited vulnerability involved a narrow technique for prompting the model to review existing minor software defects in code, emphasizing that similar outputs can be easily obtained from competing models like OpenAI without any special avoidance measures.

Anthropic's language in the statement was unusually direct: "We are complying with the government's legal directive and removing access to Fable 5 and Mythos 5 for all users. However, we do not agree that discovering a narrow potential jailbreak vulnerability should constitute grounds for recalling a commercially deployed model that has been made available to millions." The company warned that if this standard were applied across the industry, "it would essentially halt all new model deployments by advanced model providers."

The company also noted that, to date, the government has only provided verbal evidence, without written documents, technical reports, or formal disclosures through existing channels, and promised to release further technical analysis to clarify the limited nature of the issue within 24 hours.

Immediate Response in the Pre-IPO Market, Hyperliquid Perpetual Contract Plummets

This regulatory news has become a financial event because Anthropic has yet to go public, while the crypto market had already set up a "shadow pricing" venue for it.

According to CoinDesk, Anthropic's perpetual contract on Hyperliquid fell about 3.7% to approximately $1627 on Saturday, having traded above $1800 over the previous few days following the release of Fable 5. The contract's open interest was around $8.6 million, still small compared to larger private market proxy tools, but represents a notable signal for a company that has not yet filed its IPO application.

The price of this contract roughly corresponds to an implied valuation of Anthropic in the trillions—$1638 is approximately equivalent to an implied valuation of $1.638 trillion. Within 24 hours of the news breaking, the contract's 24-hour trading volume was around $190,000, with a funding rate of 0.0056%.

The impact extends beyond the contract market. According to Bitcoin.com, the ANTHROPIC token issued by Prestocks on the Solana chain, claiming to provide spot-like exposure through a special purpose vehicle (SPV) structure, dropped by 9.11% to $655.32 within 24 hours, with the related liquidity pool shrinking by 18.09% to about $126,000. It should be noted that Anthropic had previously publicly warned that any unauthorized tokenization or derivatives claiming to provide equity exposure could be considered invalid, and the aforementioned products have not been endorsed by or associated with Anthropic.

Models Halted Just Days After Release, Fable 5 Was a Milestone

The two models that were halted were just launched a few days ago.

According to reports from various media, Fable 5 and Mythos 5 were announced around June 9 and are considered a significant leap in capabilities in software engineering, scientific reasoning, visual tasks, and long-context processing. Notably, Fable 5 is the first public release of a Mythos-level model from Anthropic—this level is positioned above the original flagship Opus. The two models are built on the same underlying architecture, with the main difference being in output control: Fable 5 has a built-in classifier that intercepts responses in high-risk areas, such as cybersecurity, while Mythos 5 is open to a trusted audience subject to separate review.

According to MarkTechPost, Fable 5's classifier is an independent AI system used to identify potential misuse; when queries touch on areas like cybersecurity, biochemistry, or model distillation, it reverts to Claude Opus 4.8 for processing, with users notified of the fallback each time, triggered in less than 5% of queries. It is worth mentioning that Mythos 5 had previously supported crucial cybersecurity defense work through projects like Project Glasswing—this has also raised controversy regarding whether the "temporary disablement actually slows down infrastructure protection."

The Backlash of Security Narrative Introduces New Variables to the IPO Process

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

TechCrunch reported on this matter under the title "Anthropic's Security Warning May Have Just Backfired"—the company, which has built its entire corporate identity on "responsible AI development," is now required to shut down its strongest model based merely on a verbal briefing, which is indeed a hard pill to swallow, and the company has openly expressed this dissatisfaction.

This turmoil coincides with Anthropic advancing its IPO preparations. The company has recently secretly submitted registration documents and is among the major AI players planning to go public. Despite tools like Hyperliquid having thinner liquidity, the decline in the contract reflects investors' cautious sentiment regarding its recent valuation and path to IPO. Such sudden regulatory actions add new variables to an IPO process already fraught with uncertainty.

Anthropic stated that it is actively engaged in communication to resolve this issue and restore services as soon as possible, considering the directive to be "based on misunderstanding." The ultimate outcome of this incident may influence the future interaction between regulators and innovators and shape how export rules and security assessments apply to the continually evolving AI technologies.