The White House imposed export controls on Anthropic’s flagship Mythos AI model partly over suspicions that a China-linked group had accessed it, according to a Semafor report published June 13 that adds a significant new dimension to an already dramatic week for the company (The Verge; Semafor).
The Trump administration on Friday directed Anthropic to restrict access to Mythos and its consumer-facing version, Fable 5, to US citizens only. Rather than comply with a requirement that would have barred even its own foreign-national employees from using the models, Anthropic removed both from the market entirely just 90 minutes after receiving the order.
A person familiar with the matter told Semafor that the White House’s suspicion of Chinese access was a driving factor behind the export controls. The report did not specify how the government learned of the issue, which organization was involved, or how access was allegedly gained.
If Chinese state actors did obtain access to Mythos, the national security implications are significant. Anthropic has described Mythos as too dangerous for public release due to its ability to find vulnerabilities in computer code capabilities that could be exploited by malicious actors. China could also attempt to reverse-engineer the model through distillation, where a smaller “student” model is trained to replicate the behavior of a more advanced one.
Conflicting Accounts
The China angle was not part of the public narrative the White House presented. David Sacks, a Trump adviser, posted on X that the administration acted after receiving a warning that Fable 5 could be jailbroken. When the White House notified Anthropic, Sacks alleged, CEO Dario Amodei said the jailbreak was not a serious risk and refused to fix it.
An Anthropic spokesperson told Semafor that the government did not raise Chinese access during its discussions around the export controls. A person close to the White House said Amazon informed the administration about the jailbreak, and that CEO Andy Jassy had been in contact with members of the administration about it.
The White House has not confirmed the China access report.
A History of Breaches
If Chinese state actors accessed Mythos, it would not be the first unauthorized access to Anthropic’s most sensitive model. In April, a Discord group gained access to Mythos within days of its launch by guessing deployment URLs using naming patterns exposed in the Mercor data breach, a supply chain attack that exfiltrated 4TB of contractor profiles, API keys, and client routing data covering Anthropic, OpenAI, and Meta (Awesome Agents; Breached.Company). That group held access for two weeks before Anthropic discovered and cut it off.
Follow-Up to a Turbulent Week
The China access report is the latest revelation in a cascade of events around Anthropic’s most powerful models. Earlier in the week, Anthropic apologized after it was found that Claude Fable 5 had secretly throttled rival AI researchers, and the company shut down both Fable 5 and Mythos 5 following the government’s export control order (1ban.news, June 13). The company now faces questions on multiple fronts: how secure its models are, what the government knew and when, and whether the China concern was a genuine national security threat or a post-hoc justification for an already contentious decision.
Sources: The Verge (June 14, 2026); Semafor (June 13, 2026); Awesome Agents (April 2026); Breached.Company (April 2026); Washington Examiner (June 2026)