The U.S. White House recently imposed an export ban on the prominent AI unicorn Anthropic, citing "national security," restricting the flow of its powerful Mythos and Fable models overseas. However, foreign media columns have pointed out that, reviewing past failed attempts to regulate encryption technology (Crypto Wars) and spyware, using export bans to "contain" cutting-edge AI development is not only impractical but may ultimately undermine America’s own technological competitiveness.
(Prior context: Unblock Fable and Mythos immediately! Anthropic pledges to align closer with the White House, proposes settlement)
(Background: Musk: Chinese large models will catch up to Anthropic's Fable by Q1 2027! Zhipu's Tang Jie: It won't take that long.)
The geopolitical struggle in the field of artificial intelligence (AI) has intensified once again. According to a recent article published by senior journalist Lorenzo Franceschi-Bicchierai on June 19, 2026, the U.S. White House unexpectedly took strong action last Friday, citing "national security" as the reason to urgently order the prominent AI lab Anthropic to fully restrict the export of its top AI models, Fable and Mythos, to any entity outside the United States, including foreign nationals within the U.S.
Shortly after receiving the notice, Anthropic had its access to these two major models cut off overseas within just 90 minutes, and the restriction has now been in place for over a week. This marks the first time the U.S. government has practically tested using "export controls" to contain cutting-edge AI technology—a move that will not only impact Anthropic’s international strategy but also serve as a bellwether for global AI regulation.
Trigger for the ban: Notification from South Korean telecom providers and Amazon
What exactly caused the U.S. government to take such drastic action to ban it? According to reports, this ban was primarily triggered by two events. First, Anthropic previously granted access to Mythos through a limited partnership program to a South Korean telecommunications company (reportedly SK Telecom); however, U.S. officials suspected the company might have some connection to China, despite the telecom provider’s firm denial, which nonetheless triggered American sensitivities.
Second, Amazon CEO Andy Jassy notified the U.S. government that its internal researchers had discovered a vulnerability bypassing the security protections of the Fable 5 model. Although Anthropic emphasized that this was only a narrow, patched issue—not a full-scale "jailbreak"—it prompted the Department of Commerce to immediately impose restrictions. In fact, since launching Mythos in April of this year, Anthropic has positioned it as a "doomsday cyber weapon," restricting access to approximately 150 rigorously vetted organizations to "put the weapon in defenders' hands first."
Historical lesson: The failure of regulating cryptocurrency wars and spyware
However, is this reliance on "export controls" to contain emerging dual-use technologies truly effective? The article’s author bluntly states that historical experience offers a negative answer.
The most famous case was the "Crypto Wars" of the 1990s, when the U.S. government classified encryption technologies like PGP as weapons, fearing that intelligence agencies would be unable to intercept communications, and even launched a criminal investigation against developer Phil Zimmermann. In response, Zimmermann chose to publish the source code as a physical book, ultimately forcing the investigation to be dropped and paving the way for the end-to-end encryption era now popularized by Signal and WhatsApp. This is regarded as the most famous failure in export control history.
Another case is spyware in the 2010s. Although many Western countries classified surveillance software as dual-use technology and required export licenses under the Wassenaar Arrangement, this control became ineffective due to some countries not participating (such as Israel), lax enforcement by participating countries (such as Italy, which once permitted exports to repressive regimes), or companies relocating their headquarters to countries with weaker regulations (such as Saudi Arabia).
Banning AI exports may backfire and undermine U.S. competitiveness
The article concludes by warning that the current standoff between Anthropic and the Trump administration remains unresolved. The U.S. government faces a difficult dilemma: if it concedes and lifts restrictions, it effectively acknowledges that competitors like China will inevitably catch up; but if it insists that all future U.S. AI model exports require strict government approval, it will significantly increase compliance costs for companies and severely undermine the international market share and profitability of U.S. AI firms.
Given the historical track record of failing to control encryption technologies and spyware through strict regulations, harsh export controls have never been the right solution to prevent malicious actors from abusing powerful technologies.
