The most consequential AI story of the year for digital assets did not come from a hack, a halving, or a token launch. It came from a letter sent on a Friday afternoon.
The tweet announcing the removal of Mythos and Fable, Source: X
At 5:21pm ET on June 12, Anthropic received an export control directive from the US government instructing it to suspend all access to its two most powerful models — Fable 5 and Mythos 5 — by any foreign national, inside or outside the United States, including the company’s own foreign-national employees. The practical effect was total: to comply, Anthropic disabled both models for every customer on the planet within hours. Access to the company’s other models, including Opus 4.8, was unaffected.
According to Axios, the order came as a letter from Commerce Secretary Howard Lutnick to Anthropic chief executive Dario Amodei, citing national security authorities. An administration official told the outlet the Commerce Department moved after another company claimed it had found a way to “jailbreak” Mythos — and that the government had earlier tried, and failed, to convince Anthropic to delay the launch entirely.
For most of the technology press, this is a story about regulatory overreach and AI governance. For crypto, it is something sharper. Fable 5 and Mythos 5 are not chatbots. They are the most capable vulnerability-hunting machines ever released, and the DeFi industry had spent the previous week arguing about whether their arrival was a gift or a death sentence. Now Washington has answered the question by taking them away — from the white hats as well as the black hats.
Users can no longer use Fable, source: Claude
What the government says, and what Anthropic says back
The two sides do not agree on much. Anthropic’s account is unusually blunt for a company complying with a federal order. It says the government provided no specific details of its national security concern in the letter itself, and that the underlying issue appears to be a single, narrow “jailbreak” — one that essentially amounts to asking the model to read a codebase and fix its flaws. The company says it reviewed a demonstration of the technique being used to surface a handful of previously known, minor vulnerabilities, all of which other publicly available models can find without any bypass at all.
Anthropic went further, arguing that no tester has yet found a universal jailbreak capable of broadly unlocking the model’s blocked capabilities, and that the narrow findings disclosed so far produced “either entirely benign responses or are minor findings that provide no Mythos-specific uplift.” It pointed to OpenAI’s GPT-5.5 as offering comparable capability, and warned that if a single narrow jailbreak were grounds to recall a model deployed to hundreds of millions of people, “it would essentially halt all new model deployments for all frontier model providers.”
The company says it is complying while disputing the basis, calling the episode a “misunderstanding” and promising to restore access as soon as possible. That is the corporate position. The market does not wait for corporate positions.
Why crypto was watching this model in particular
To understand why this matters for digital assets, recall what these models can do. Mythos Preview — the research-grade ancestor of the now-suspended pair — found a 27-year-old vulnerability in OpenBSD, surfaced critical flaws across more than 1,000 open-source projects including the Linux kernel and FFmpeg, and, in one widely circulated example, identified a critical bug in the Zcash protocol within 24 hours — a flaw that had survived four years of scrutiny from some of the world’s best cryptographers.
Crypto’s architecture makes it uniquely exposed to a tool like this. Traditional finance runs on siloed, proprietary systems with circuit breakers and centralized fail-safes. DeFi runs almost entirely on public code: open-source dependencies, browser wallets, RPC infrastructure, and smart contracts that are transparent to anyone — human or machine — who wishes to read them. An AI that can find ancient bugs in hardened operating systems is, in principle, an AI that can find the unaudited reentrancy flaw sitting in a protocol’s contracts right now.
The defenders lose a tool, too
Here is the awkward part. The same models that worried the attack-side of the ledger were rapidly becoming infrastructure on the defense side. NYSE and ICE had begun deploying Mythos for cybersecurity; exchanges including Coinbase and Binance, and DeFi teams such as Uniswap, had sought early access. Anthropic’s own Project Glasswing partners reported fixing hundreds of vulnerabilities with the model’s help — Mozilla alone among them.
Suspend the model and you do not simply disarm a hypothetical attacker. You also pull the audit tool out from under every protocol and security firm that had started to rely on it. The asymmetry is unforgiving: a motivated adversary who already extracted what they needed, or who can run a comparable open-weight model, keeps their capability; the compliant auditor who used Fable 5 through the front door loses theirs overnight.
That points to the deeper systemic risk that predates this directive: concentration. As one analysis noted before the suspension, if a meaningful share of critical financial infrastructure comes to depend on the same model from the same provider, a flaw — or, as it turns out, a government order — affecting that one model becomes its own single point of failure. The crypto industry spent a decade learning to distrust single points of failure. It is now discovering that its newest security dependency had a kill switch in Washington.
The market shrugged, for now
Crypto markets, preoccupied with macro, barely registered the news. Bitcoin was trading around $63,400 on Friday, up modestly on softer core CPI data and optimism around a possible US-Iran de-escalation, with the Fear & Greed Index still pinned in “extreme fear.” Ether held near $1,670. There is no clean, immediate price impact from an AI export control — and that is precisely the point. The risk here is not a candle on a chart; it is the slow re-rating of an assumption the industry had quietly adopted, that frontier AI would be a permanent, dependable fixture of the security stack.
For now, the live question is regulatory, not technical. Anthropic has argued for years that governments should be able to block unsafe deployments — but through a process that is transparent, fair, and grounded in technical facts. It says this action met none of those tests. Whether one reads that as principled objection or commercial self-interest, the precedent is now set: a single agency letter can remove a deployed frontier model from global circulation in an afternoon.
Crypto, of all industries, understands what it means when access depends on the goodwill of an authority that can change its mind. It just did not expect the lesson to arrive via its AI vendor.
The backstory: a year of escalation
Friday letters have become something of a motif in Anthropic’s dealings with this administration. The first arrived on February 27, when — after a months-long standoff with the Pentagon and a 5pm deadline the company let pass — President Trump ordered every federal agency to stop using Anthropic’s products, and Defense Secretary Pete Hegseth moved to designate the company a “Supply-Chain Risk to National Security,” a label normally reserved for foreign adversaries, barring defense contractors from doing any business with it. Hegseth’s framing was characteristically blunt: “America’s warfighters will never be held hostage by the ideological whims of Big Tech.” The dispute’s root cause was Anthropic’s refusal to drop two conditions — no fully autonomous weapons making final targeting decisions, and no use of Claude for mass domestic surveillance.
Anthropic did not blink. It called the designation unlawful, argued the Pentagon’s authority could not extend to how non-defense customers use Claude, and on March 9 filed two federal lawsuits alleging violations of the Administrative Procedure Act and its due-process and First Amendment rights. Microsoft, an investor and close partner, filed an amicus brief in support — a notable show of Big Tech solidarity against a sitting administration.
The contradictions only deepened. On June 5, Trump signed National Security Presidential Memorandum 11, directing agencies to accelerate AI adoption and to terminate contracts with any AI firm that “repeatedly” limits government use — a clause widely read as aimed squarely at Anthropic. Yet the same memo carved out an exception for Claude Mythos, which was already deployed at the NSA, granting it a waiver. The administration was simultaneously trying to push Anthropic out of government and quietly keeping its most powerful model on for national-security work.
That is the throughline worth holding onto. It was precisely Mythos’s ability to find and exploit cyber vulnerabilities that pushed a deregulation-minded White House to embrace AI oversight tools it had previously dismissed. Friday’s export control directive is not a bolt from the blue. It is the latest move in a year-long fight in which the government has alternately tried to ban Anthropic, sue-proof its own conduct against Anthropic, and depend on Anthropic’s models — sometimes in the same week.
