Anthropic told customers on Friday night that it would suspend access to its two latest models, Fable 5 and Mythos 5, after the U.S. government invoked export controls on the products. The Department of Commerce informed the AI lab that it had learned of a method to "jailbreak" Fable 5, allowing users to sidestep the safety limits the company built into the system. In response, Anthropic shut the models down entirely, a step that also bars any foreign national—whether inside or outside the United States—from using them.
When Anthropic first unveiled Mythos, it limited the rollout to a handful of government agencies and technology professionals because of the model’s ability to uncover cybersecurity flaws. The sudden restriction, however, forces the company to pull the models from all users, including its own foreign‑national staff, to comply with the export‑control order. "We disagree that the jailbreak shown warrants such sweeping action," the company wrote in a statement, warning that a precedent like this could "essentially halt" the development of future frontier AI models.
Anthropic’s leadership has already flown to Washington to discuss the matter with senior officials. According to a report in the Wall Street Journal, company executives spent several hours on calls Saturday with Secretary of Commerce Howard Lutnick and National Cyber Director Sean Cairncross. A company spokesperson declined to add further comment beyond the written statement. The Department of Commerce has not responded to requests for comment.
The clash marks the latest showdown between the AI lab and the federal government. Earlier this year, the Pentagon labeled Anthropic a "supply chain risk" after the firm refused to allow its Claude models to be used for fully autonomous weapons or mass domestic surveillance. Defense officials wanted unrestricted access for "all lawful purposes," a demand the company resisted.
David Sacks, co‑chair of the President's Council of Advisors on Science and Technology and former "AI and crypto czar" under the Trump administration, dismissed any suggestion that the current dispute is linked to the Pentagon issue. In a lengthy post on X, Sacks said the administration believes the problem is "serious but should be easily resolved" and that "the ball is in Anthropic's court." He added that the company’s reputation for safety is at stake, noting that Anthropic appeared to prioritize keeping a consumer model available over addressing the security gap.
Experts say jailbreaks are an expected challenge for advanced language models. Ayham Boucher, executive director of AI strategy and innovation at Cornell University, explained that any sufficiently capable model can be coaxed into disallowed behavior through role‑playing or other prompt‑engineering tricks. "All models can be jailbroken," he said, emphasizing that the battle between developers and adversaries is ongoing. Anthropic had built a 30‑day data‑retention policy into Fable 5 to monitor interactions and detect potential exploits—a safeguard that may have limited its appeal to privacy‑sensitive customers.
While Anthropic is the focus of the current controversy, the broader AI community is watching closely. Researchers note that Mythos’s coding and cybersecurity prowess is comparable to OpenAI’s upcoming GPT‑5.5, and that other firms—both U.S. and Chinese—are racing to achieve similar capabilities. Boucher warned that the export‑control action could slow model development not only at Anthropic but across the industry, especially if foreign talent is barred from contributing to future releases.
The Trump administration, which has framed AI dominance as a national security priority, signed an executive order earlier this month urging developers to submit frontier models for federal review before release. Critics argue that the order’s voluntary language has already eroded, as seen in Friday’s abrupt shutdown. Samir Jain, vice president of policy at the Center for Democracy and Technology, described the move as "arbitrary and opaque," raising questions about the legal basis for the controls and potential First Amendment implications.
For now, Anthropic’s customers are left without access to two of the most advanced tools in the market. The company has indicated it will continue negotiations with Washington, but the outcome remains uncertain. The incident underscores the growing tension between rapid AI innovation and the government’s effort to mitigate security risks, a balance that will shape the industry’s trajectory in the months ahead.
Este artículo fue escrito con la asistencia de IA.
News Factory APP - noticias agénticas para impulsar tu SEO y AEO.