The Trump administration lifted the ban on Anthropic's most powerful AI model, Fable 5, but regulatory uncertainty still needs to be eliminated.

The US government has lifted the access restrictions on Anthropic's Fable 5 artificial intelligence models. After the AI startup resolved security concerns raised by the Trump administration, the models were approved for broader release. On June 30th, the US Department of Commerce sent a letter to Anthropic co-founder Tom Brown officially announcing the removal of export controls on Fable 5 and Mythos 5. Anthropic confirmed receipt of the notification from the Department of Commerce on X platform on Tuesday, stating that access would be restored the following day. The company said, "We thank users for their patience and appreciate all those who helped us redeploy these models." Previously, Mythos 5 had been granted limited access to certain US businesses and government agencies last Friday. The US government's latest measures further expand the access rights of Fable 5 to ordinary users. Fable 5 is positioned as a consumer product, a public version of Anthropic's powerful Mythos 5 model with additional security mechanisms aimed at preventing the software from being used for launching cyber attacks. On June 9th, amidst much anticipation, Anthropic released its groundbreaking Mythos-level large model. Fable 5, as the first Mythos-level product open to the public, outperformed competitors in almost all benchmark tests. It scored 80.3% in the SWE-Bench Pro programming test, far surpassing GPT-5.5's 58.6%. In the Frontier Code Diamond test, which is closer to real-world engineering scenarios, it left GPT-5.5 far behind with a score of 29.3% compared to 5.7%. This is not just a competition between technologies of the same generation but a crushing of intergenerational differences. However, just 3 days later, citing national security reasons, the US government placed export controls on Fable 5 and Mythos 5 models. This control went beyond conventional export restrictions, not only prohibiting access to individuals and organizations outside the US but also including foreign nationals within the US, including foreign employees at Anthropic, in the ban list. As it was technically impossible to quickly and accurately determine the nationality of each user, Anthropic ultimately chose to "cut it all off", fully suspending the two new models and reverting all user requests to the previous versions. Other models under the company were unaffected. Simultaneously, the company has been negotiating with government officials to address their concerns. In recent days, top executives at Anthropic, including Tom Brown, have been in talks with US Secretary of Commerce Howard Lutnick and other government officials to address the government's concerns regarding the company's latest model release. The key to lifting the export controls lay in Anthropic successfully eliminating concerns from White House officials regarding the risk of "Jailbreak" by malicious actors. In the field of AI, "Jailbreak" refers to bypassing the built-in security features and content restrictions of an AI system using carefully designed prompts or technical means to generate prohibited content or carry out restricted operations. For regular AI models, "Jailbreak" may only result in generating inappropriate or false information. But for Mythos-level models like Fable 5, the consequences of "Jailbreak" are on a completely different level. As noted by David Sachs, co-chair of the President's Council of Advisors on Science and Technology (PCAST) and former White House AI Policy Lead, "Fable is essentially a Mythos with guardrails, once those guardrails are breached, it essentially exposes Mythos and its advanced capabilities for network attacks to unauthorized individuals." According to a letter from Lutnick to Anthropic on Tuesday, Anthropic committed to "proactively mitigate and resolve security risks associated with these models." Lutnick mentioned that Anthropic also agreed to collaborate with the US government to establish relevant "protocols and standards" for future models. It is worth mentioning that to facilitate the lifting of restrictions, Anthropic noticeably adjusted its position and communication strategy in recent negotiations. The company previously insisted that the US government's concerns about the Jailbreak risk were exaggerated and expressed an inability to guarantee the limited capabilities of the Mythos model. However, according to sources, Anthropic subsequently shifted stance, pledging to reduce the occurrence of "Jailbreak" incidents by building stronger security measures. Analysts believe that this essentially aligns with the demands of the US government, no longer arguing about the conceptual issue of whether "Jailbreak" can be eradicated. In addition, at the internal communication level, Anthropic CEO Dario Amodei has recently stepped back from direct negotiations with the government, with co-founder Tom Brown taking over meetings. Trump administration officials are said to be more accepting of Brown on a personal level. Although the export controls on Anthropic models have been lifted, the order issued by the US Department of Commerce represents the most significant intervention by the US government in an AI company's operations to date, sparking legal disputes on whether export controls could be used to regulate access to AI models. Despite the lifting of restrictions on Fable 5 and Mythos 5, concerns persist regarding the uncertainty surrounding US government AI regulations. The US government's decision to restrict the availability of Anthropic models has triggered a chain reaction in the entire AI industry, particularly affecting competitors like OpenAI. Last week, OpenAI's ChatGPT-5.6 model, at the request of the White House, the Office of Science and Technology Policy, and the Office of the National Cyber Director, was only made available to a small group of users, despite CEO Sam Altman publicly expressing displeasure with this release method. Senior executives at OpenAI expressed hope that an executive order aimed at evaluating the most advanced AI models could be implemented soon, establishing a standardized process for making such decisions to curb the Trump administration's temporary and unpredictable actions in AI regulation. This call has resonated throughout the industry. The ongoing game between the government and the industry over AI regulatory frameworks is likely to continue in the short term.