AP Photo/Patrick Sison
The U.S. government has ordered a suspension of access to the advanced AI models Claude Fable 5 and Mythos 5 for all foreign nationals, whether in the country or abroad, including overseas employees of Anthropic, the company that developed them. According to the firm, the decision was based on cybersecurity concerns that officials believe threaten national interests.
An Anthropic spokesperson noted that they have temporarily disabled Claude Fable 5 and Mythos 5 for all clients to comply with the requirements. The company stated that the government provided «only verbal evidence» of potential narrow and unproven hacking methods that involve querying the model to check for errors in its underlying code.
The developers at Anthropic examined a demonstration of this specific method, which was used to identify a few previously known minor vulnerabilities. According to Anthropic, all of these vulnerabilities are «relatively simple,» and other «public models» can also detect them «without needing to bypass the system.»
“We disagree that discovering a potential narrow loophole in the security system should justify withdrawing a commercial model used by hundreds of millions of individuals,” the company stated.
Anthropic clarified that no AI model provider can guarantee complete immunity to so-called jailbreaks—bypassing the protective measures of models.
Furthermore, when releasing Fable 5 in early June, developers acknowledged that universal jailbreaks might be discovered in the future. In the weeks leading up to the model’s launch, collaborative vulnerability testing was conducted with the U.S. government, the UK’s Cybersecurity Agency, several private third-party organizations, and internal teams.
According to company representatives, these tests showed that the protective measures of Fable are significantly more effective than any previously used models. No tester was able to find a universal method to bypass the protection during evaluations.
“We have implemented robust protective measures that significantly reduce the likelihood of Fable being misused for cybersecurity-related tasks. In fact, our safeguards are so effective that many users complain about their excessive stringency,” Anthropic emphasized.
The company has apologized to clients for the situation and is working on restoring access. Although both models were built on the same technical foundation, only Fable 5 was publicly presented. It imposes much stricter limitations on the types of questions users can ask, particularly on topics related to cybersecurity and biology, NBC News reports.
Mythos 5, on the other hand, was introduced without such restrictions to a select group of trusted partners, including key companies in the cybersecurity and infrastructure sectors, the channel adds. As previously reported by RTVI.US, the capabilities of Mythos have seriously alarmed not only the United States but also China. The AI is capable of hacking into state databases, financial systems, and medical networks, exposing their vulnerabilities.
Anthropic’s decision to suspend user access to the platforms marks the first instance where a leading AI company has disabled a publicly available model due to federal government intervention, NBC News notes.
The actions of U.S. authorities toward Anthropic reflect a growing tendency in Washington to view advanced AI systems as tools for national security, Axios reports. According to the publication, U.S. Secretary of Commerce Howard Lutnick sent a letter to the company’s CEO Dario Amodei, informing him that the models will be subject to export control.
This is the second instance where the Trump administration has taken action against this tech company, The New York Times (NYT) reports. The paper mentioned “tense negotiations” in March regarding how Anthropic’s AI might be utilized in military and intelligence contexts.
The Pentagon stated at the time that it viewed the corporation as an “unacceptable source of risk,” which prevents federal agencies from using its tools, the publication observes.