Export restrictions imposed by the White House on Anthropic's artificial intelligence models were a direct consequence of concerns about potential access to the flagship Mythos model by a group linked to China. This decision, made in mid-June, caught the industry off guard, as such measures had previously been applied only to semiconductor technology, not software.
An additional trigger was reports of an allegedly discovered method to bypass the protections of the Fable 5 model, which was positioned as a safe version for broad access. Although Anthropic itself insists that it was a narrow and non-universal method that did not grant the model unique capabilities, authorities deemed the threat serious enough.
Timeline of Events and Positions of the Parties
According to my analysis, the key moment was the interaction between the Donald Trump administration and Amazon's leadership. On June 11, Amazon CEO Andy Jassy notified the White House of a potential way to bypass Fable 5's restrictions. This occurred as part of a model review requested by authorities. Subsequently, U.S. Secretary of Commerce Howard Lutnick sent a letter to Anthropic CEO Dario Amodei, mandating export controls for both foreign individuals within the United States and those abroad.
The administration's tactics are interesting: officials tried to persuade Anthropic to voluntarily suspend access, but the company refused. After that, the startup was given just 90 minutes to disable the models without any details about the nature of the threat. White House advisor David Sacks later stated that the warning came from a trusted partner whom both sides believe.
Anthropic's Position and Consequences
Anthropic strongly disagrees with the threat assessment. The company claims that the authorities' letter contained no specific details about risks to national security. The startup conducted its own investigation and concluded that the demonstrated technique only allowed finding "a small number of previously known minor vulnerabilities" that other publicly available models can also identify without bypassing protections.
Let me remind you that before launching Fable 5 and Mythos 5, the company spent "thousands of hours" testing alongside the U.S. government and the UK's AI Safety Institute. Anthropic views the situation as a "misunderstanding" and is working to restore access. Notably, amid these events, tokens of decentralized AI projects Venice and Morpheus rose by 14% and 21% respectively—the market clearly saw centralized restrictions as an opportunity for alternative solutions.
My comment: This precedent demonstrates that regulators are beginning to perceive advanced AI models as a strategic asset, akin to nuclear technology. However, the haste and lack of transparency in the administration's actions create a dangerous precedent where decisions to block entire classes of technology can be made based on unverified data. This undermines market confidence and stimulates the development of decentralized and censorship-resistant alternatives.