Anthropic locks new top AI model at the behest of the US government

Fable 5 had been released earlier the same week and is based on the underlying Mythos technology, whose cybersecurity and biotechnology capabilities are deactivated in the public version. According to Anthropic, the non-public full version Mythos 5 will continue to be used only by government agencies and selected corporate partners to harden their systems. "Mythos 5 ist die nicht-öffentliche volle Version, die weiterhin nur von Behörden und ausgewählten Unternehmenspartnern zur Härtung ihrer Systeme eingesetzt werden sollte."

What Fable 5 and Mythos 5 are about

The background to the intervention, according to the company, is a report that Anthropic reviewed and which the firm believes likely triggered the government order. "Der Firma zufolge glaubt die Regierung, Kenntnis von einer Methode erlangt zu haben, Beschränkungen in der Software aufzuheben." The company's own experts, however, reached a different conclusion. "Dabei seien die Anthropic-Experten zu dem Schluss gekommen, dass es sich dabei um eine eingeschränkte Möglichkeit handele, die KI spezifische Programmcodes prüfen und Fehler korrigieren zu lassen."

Anthropic is pushing back against the blanket block. The company told the government it disagreed "dass deswegen Software für Hunderte Millionen Nutzer blockiert werde". It has also so far only received partial information from the government. "Anthropic betonte, man habe bisher nur teilweise Informationen von der Regierung erhalten."

The company also points out that comparable capabilities exist in competing products. "Diese Fähigkeit hätten auch Modelle anderer Anbieter wie etwa GPT-5.5 des Rivalen OpenAI." Anthropic emphasizes that the safety precautions in Fable 5 have been extensively tested. "Die Firma betont, dass die Sicherheitsvorkehrungen bei Fable 5 ausgiebig getestet worden seien."

Anthropic's criticism of the government's approach

The company sets out fundamental requirements for such a move: any block on AI software must be based on transparent and clear procedures as well as technical facts – which, in Anthropic's view, is not the case here. The exceptional cybersecurity capabilities from the Mythos model had made the company increasingly important to the US government in recent months: "Die Künstliche Intelligenz hinter Anthropics KI-Modell Mythos ist besonders gut darin, zum Teil auch über Jahrzehnte unentdeckt gebliebene Software-Schwachstellen aufzuspüren." So far, this capability has been used by US agencies and selected companies "um die Sicherheitslücken zu stopfen".

Background: conflicts between Anthropic and Washington

The conflict with Washington is not new. In recent months, Anthropic has repeatedly made headlines due to disputes with the US government. The Pentagon had classified Anthropic as a supply-chain risk, which can significantly hinder the use of the software in government agencies; the company is suing over this classification. Despite the pressure, Anthropic has also insisted that its models not be used in autonomous weapons systems or for mass surveillance in the United States.

Notable is the contrast with the position of Anthropic CEO Dario Amodei, who only days earlier had argued that the government should be allowed to block dangerous AI software. Now the company itself is carrying out such a block – albeit under protest and with the note that the chosen procedure was non-transparent and decided over the heads of hundreds of millions of users.