Washington Shows Anthropic and the AI Industry What It Can Do to Them

\ Is the Trump administration making an example of Anthropic by restricting access to the company’s most advanced AI models? The question arose almost immediately after Anthropic announced on June 12 that it was disabling access to Fable 5 and Mythos 5 for all customers. According to the company, the decision followed a directive from the U.S. government requiring licenses for exports, re-exports, and transfers involving the models to foreign nationals. Anthropic said the order left it little choice but to suspend access entirely while it sought clarification. Anthropic argued that the U.S. government had relied on limited evidence and overstated the risks posed by the models. The directive came only days after Anthropic introduced Fable 5 and Mythos 5 and highlighted their cybersecurity capabilities. It also happened shortly after the Trump administration outlined a broader strategy aimed at preserving American leadership in AI while treating advanced capabilities as matters of national security. There is good reason to believe Anthropic is being singled out given Anthropic’s contentious relationship with the Trump administration . At the same time, with the June 12 development, Washington has also shown that it is willing to reach into a private AI company and restrict access to its most advanced models. OpenAI, xAI, Google, and every future frontier AI lab now have to assume it could happen to them too. \ Why Anthropic? Anthropic has spent years arguing that frontier AI deserved extraordinary treatment. CEO Dario Amodei repeatedly warned about biological threats, cyberattacks, and the possibility that future systems could exceed human capabilities in important domains. If the government eventually accepted those arguments and turned them back on Anthropic, the irony is difficult to miss. Anthropic also chose to introduce Mythos 5 as the strongest cybersecurity model in the world. That may have sounded like a marketing message inside Silicon Valley, but the Trump administration may have heard something different. Anthropic may have given Washington the ammunition it needed to restrict access. Anthropic and the Trump administration were already divided over the role of AI in national security. The company challenged parts of the administration’s agenda and clashed with the Department of Defense over military applications. Dario Amodei became one of Silicon Valley’s most prominent advocates for treating frontier AI as a source of catastrophic risk. OpenAI spoke the language of competition and American advantage . The directive hit days after the Trump signed an executive order establishing a voluntary framework under which AI developers could provide the government with early access to frontier models before releasing them to other partners. Anthropic expanded access to Mythos on the same day the order was signed. Fable 5 launched without going through the pre-release review process the administration had just formalized. The government had stated its intentions publicly, and Anthropic proceeded anyway. \ What Comes Next? The government’s June 12 action cited a narrow, non-universal jailbreak, a technique that could extract some cybersecurity information in specific circumstances but could not broadly bypass the model’s safeguards. Anthropic argued that other publicly available models, including OpenAI’s GPT-5.5, produce equivalent results without any bypass. GPT-5.5 has a cybersecurity variant. Weeks ago, OpenAI announced GPT-5.5-Cyber as a frontier cybersecurity model for critical cyber defenders. OpenAI noted that early testing partners used it to automate red-teaming exercises on infrastructure and validate high-severity vulnerabilities. Apparently OpenAI’s approach was acceptable to the same government that pulled Fable 5 from the market for doing something similar under a jailbreak. OpenAI restricted access and built government collaboration in from the start. In context of what happened on June 12 (and what did not happen to OpenAI), Trump is sending a message: play by my rules, or else. Image source: Steve A. Johnson, Unsplash \