Red Hot Cyber
Cybersecurity, Cybercrime News and Vulnerability Analysis
As expected, Anthropic’s response to the blockade of the Mythos 5 and Fable 5 AI models did not hesitate to arrive. Already on Monday, the company met with representatives of the department to initiate negotiations on the matter in question.
Informed sources about the events have informed Bloomberg of the start of negotiations. We recall that at the end of last week, the US government expressed concern about the threat related to the escape of cutting-edge AI models for national security issues.
Consequently, the US government ordered Anthropic to block access to these artificial intelligence models for foreign citizens, even if they are in the United States. Not seeing the possibility of providing individual access blocking, Anthropic simply disconnected all users from Mythos 5 and Fable 5.
Previously, these models were distributed under specific controls, taking into account their ability to quickly find security bugs in IT infrastructures. The government ordinance represents a real challenge on many levels of analysis, such as business for the Anthropic company, the use of such models, and also from the strictly related point of national sovereignty and the use of the cloud.
After the ban, a significant part of the US cybersecurity community asked the White House to review its positions, which led to the blockade of Anthropic’s most advanced models. They argue that the measure risk penalizing the defenders above all. In a letter addressed to Secretary of Commerce Howard Lutnick and the head of national cybersecurity Sean Cairncross, 40 experts and executives including top figures from Adobe, Zoom, and Sophos, reported that setting limits on access to cutting-edge models could slow down analysis, detection, and response to cyber attacks.
According to the document, the contested capabilities would concern the creation of proof of concept related to vulnerabilities. These are useful tools for both attackers and defenders, who can understand and correct the identified security gaps.
The debate was relaunched by Alex Stamos and other experts such as Katie Moussouris and Chris Wysopal, who argue that the same capabilities are already present in other advanced models, including competing systems and open-source models. For critics of the measure, the risk is that of creating a strong imbalance: limiting Western defenders while state and criminal actors continue to develop and use equivalent tools.
Bloomberg did not report who exactly took part in the negotiations. Anthropic’s representatives confirmed the meeting, emphasizing that both parties are working quickly to reach a solution to the problem. The face-to-face meeting with officials was preceded by numerous video conferences.
The Mythos model was distributed under the strict control of American officials in April of this year, and its version Fable 5 was intended for a wider audience.
As Anthropic’s representatives suggest, officials were concerned about the emergence of unauthorized access to the models’ capabilities and thus transforming them into a powerful tool for cyber attacks. Third-party experts have noted that eliminating potential vulnerabilities in Fable 5 could take a long time.
The situation is complicated by the previous contradictions between Anthropic and the US authorities. We recall that this year the Pentagon said that the company can be defined as a risk to supply chains. This happened after Anthropic’s refusal to use artificial intelligence for unethical purposes, such as combat, as well as surveillance of US citizens.
