New reporting names the two events that pushed the White House to act, while three decades of export control history suggest this ban may not hold. The outcome directly affects which AI tools non-U.S. SaaS teams can access.
- SK Telecom, a South Korean telecom widely reported to be the foreign company involved, had been granted access to Mythos through Anthropic’s limited partner program before U.S. officials flagged it as a suspected China-linked entity.
- Amazon CEO Andy Jassy reportedly alerted the White House separately after Amazon researchers identified a potential workaround in Fable 5’s safety layer. Anthropic disputes this, calling it a narrow, already-patched issue.
- Anthropic had roughly 90 minutes from notification to restrict access. Both Fable and Mythos have been offline for a week with no resolution in sight.
- Thirty years of precedent from PGP encryption and spyware regulation suggest government export controls on dual-use software consistently fail to contain the technology.
What Happened
New reporting traces two specific events that pushed the Trump White House to ban Anthropic’s Fable and Mythos models from export last week.
A South Korean telecom, widely identified as SK Telecom, had been granted access to Mythos through Anthropic’s limited partner program. U.S. officials grew alarmed after flagging the company as a suspected China-linked entity. SK Telecom has publicly denied any connection to China.
Amazon CEO Andy Jassy separately alerted the administration after Amazon researchers identified a potential workaround in Fable 5’s safety layer. Anthropic disputes the jailbreak label, describing the issue as narrow and already patched well before the ban was issued.
Anthropic had roughly 90 minutes from notification to restrict access to both models. Fable and Mythos have remained offline for a week. The standoff between Anthropic and the Trump administration is unresolved.
The vulnerability was a narrow, already-patched issue, not a wholesale defeat of the model’s safety measures. Anthropic, disputing the Amazon jailbreak claim
Why It Matters
The Mythos ban fits a 30-year pattern of governments trying to contain dual-use cyber technology through export controls. Those efforts have consistently come apart. In the 1990s, U.S. encryption controls were effectively beaten when PGP’s creator published his source code as a physical book. Later, international spyware controls under the Wassenaar Arrangement allowed companies to simply relocate to jurisdictions with looser rules, and the most sanctioned players kept operating under different names.
The skeptic argument cuts both ways. If the ban holds, it creates a compliance layer that will affect every AI lab trying to serve foreign enterprise customers, not just Anthropic. If the administration lifts it, the move tacitly confirms that other countries, including China, will reach comparable AI capabilities regardless of what the U.S. restricts. Neither outcome resolves the underlying tension.
Bottom Line
The most immediate question is whether the Trump administration lifts the ban to protect U.S. AI competitiveness abroad. Precedent from both the Crypto Wars and spyware regulation suggests software restrictions rarely hold at the border.
For SaaS teams currently using or evaluating Anthropic’s models, the practical near-term risk is compliance friction before AI vendor access, not permanent loss of access. Relve, an AI trends intelligence platform, tracks how policy shifts like this ripple into enterprise AI vendor decisions.
