China-Linked Group May Have Accessed Anthropic's Mythos. The White House Never Told Anthropic That Was the Reason.

Since the White House directed Anthropic to restrict Mythos and Fable 5 to U.S. citizens last week, forcing the company to pull both models entirely, a second explanation for the order has surfaced — one that the government never shared with Anthropic.

According to Semafor's Reed Albergotti, reporting updated June 13, the export control directive was driven in part by suspicions that a China-linked group had accessed Mythos. The identity of that group, how it gained access, and how the White House learned of the breach are all still unknown, per Semafor's sourcing.

An Anthropic spokesperson told Semafor plainly: the government did not raise Chinese access to Mythos in its conversations around the export controls. Anthropic also said it prohibits use of its products from China entirely, according to Ynet News.

Two Justifications, One Order

The public-facing rationale, stated by White House AI and Crypto Czar David Sacks in a post on X, focused entirely on something different: a jailbreak. Sacks alleged that the government received a warning that Fable 5 could be jailbroken to remove its cybersecurity guardrails, that the White House notified Anthropic CEO Dario Amodei, and that Amodei dismissed it as not a serious risk and declined to fix it. Sacks said the administration then issued the export controls reluctantly.

Anthropic denied that the jailbreak was a meaningful threat. The company said in its Friday blog post, cited by Al Jazeera, that the government provided only "verbal evidence of a potential" issue — no documentation, no technical demonstration.

Sacks' X post mentioned ZERO about China. Whether that omission was deliberate, a communications gap between White House factions, or a reflection of unconfirmed intelligence is not established in any of these sources.

What Chinese Access Would Actually Mean

If a China-linked group did access Mythos, the implications go beyond an embarrassing breach. As both The Verge and Semafor noted, China could attempt to replicate the model through a process called distillation, training a less capable "student" AI on the outputs of Mythos to approximate its behavior at a fraction of the development cost. Anthropic itself has described Mythos as too dangerous for public release because of its ability to identify exploitable bugs in computer code.

This would not be the first time Mythos was accessed outside intended channels. Anthropic previously disclosed that a Discord group had access to the model for two weeks before the company discovered the breach and revoked it, according to The Verge.

Amazon's Role

A source close to the White House told Semafor that Amazon — which is a major Anthropic investor — informed the administration about the jailbreak, and that Amazon CEO Andy Jassy had been in contact with White House officials about it. Neither Amazon nor Jassy has made a public statement confirming that account. The Verge reported the Amazon angle but neither confirmed nor disputed the characterization of Jassy's involvement.

Anthropic and Amazon have deep financial ties, making Amazon's reported role in triggering a government crackdown on its own investee company an unusual dynamic worth scrutiny. Whether Amazon flagged a genuine security concern or whether there is a competitive dimension to this has not been established by any source.

The Strongest Counterargument

Critics of the White House action have a legitimate point: the government has offered no documented, verifiable evidence to the public, or apparently even to Anthropic, that Chinese access actually occurred. Anthropic stated it was never told China was the reason for the order. A company cannot respond to an accusation it was never given. If the intelligence underlying the export controls is thin, unverified, or based on a single source, the result is a U.S. AI company having its flagship models pulled from the global market over an allegation it cannot examine or contest.

The concern is worth examining. "The government didn't tell us" is not the same as "it didn't happen." Intelligence about foreign access to sensitive technology is frequently kept compartmentalized precisely because disclosing it would reveal collection methods. The government not briefing Anthropic fully is consistent with standard counterintelligence practice — and also consistent with a sloppy, poorly communicated policy decision. Both remain possible.

Where This Sits as of June 14

As of today, three things are factually established: the export control order was issued, Anthropic pulled both models rather than comply with citizen-only access requirements, and two distinct justifications are now in circulation — a jailbreak dispute and a suspected Chinese breach — with no official statement confirming both or reconciling the two.

The unresolved question with the most practical consequence: Anthropic's lawsuit against the Trump administration over a separate supply chain blacklist is already active, according to Al Jazeera. Whether the government's decision to withhold the China justification from Anthropic directly affects that litigation — or prompts the company to seek discovery on what the White House actually knew and when — is a concrete legal question that will likely surface in coming weeks.