Israel, EU press Anthropic for access to restricted Mythos AI model

Israel’s cyber directorate chief Yossi Karadi said he is pressing AI labs including Anthropic for access to models like Mythos to defend government networks against Iranian threats.ground
The European Commission has held multiple meetings with Anthropic but acknowledged talks lag behind a deal already struck with OpenAI for its rival cyber model.reuters
Anthropic signaled it plans to expand Mythos access to allied governments and expects broad public availability within 6-12 months, per a Project Glasswing update.theregister

EU and Israel Push for Access to Anthropic’s Restricted Mythos AI Model

The European Commission and Israel’s top cyberdefense official are independently pressing Anthropic for access to its Claude Mythos model, the restricted artificial intelligence system that has upended cybersecurity debates since its limited unveiling in April. The parallel efforts underscore growing international frustration with the U.S.-centric distribution of a tool that governments say they need to defend critical infrastructure against the very attacks Mythos could enable.

Israel’s Cyber Chief Makes His Case

Yossi Karadi, director-general of Israel’s National Cyber Directorate, revealed in an interview published Tuesday that he is pressing major AI labs for controlled access to advanced models like Mythos to help defend Israeli government networks. The disclosure came alongside warnings that Iranian state-backed hackers have grown more coordinated and are increasingly using AI for influence operations since the U.S.-Israel war with Iran began.ground

Karadi has long warned of an approaching “cyber war” driven by AI agents. In a special recommendations report published this month, the INCD said models like Mythos and OpenAI’s GPT-5.4-Cyber have “broken through the technological complexity barrier,” enabling attackers to identify and exploit vulnerabilities at “machine speed”. The directorate is urging Israeli organizations to shift from a prevention-first strategy to a “Breach Ready” posture that assumes networks will be compromised.israeldefense

Europe’s Months-Long Campaign

The European Commission has been engaged in a sustained campaign to obtain Mythos access since shortly after the model’s April 7 announcement. European Economic Commissioner Valdis Dombrovskis confirmed in early May that Commission representatives had met with Anthropic and been briefed on the model’s technical capabilities. By mid-May, however, Commission spokesman Thomas Regnier acknowledged that talks with Anthropic were “not yet at the same stage” as those with OpenAI, which had already agreed to provide its GPT-5.5-Cyber model to European partners.reuters

An EU official had previously visited San Francisco in April to meet with both OpenAI and Anthropic about AI regulations. Germany’s Bundesbank has separately called on the EU to secure broader access, warning that restricted availability could leave European financial systems vulnerable.bizjournals

Anthropic Signals Expansion

The pressure appears to be yielding results. In a Project Glasswing update published last week, Anthropic said its next step would be to “work with critical partners — including US and allied governments — to expand Project Glasswing to additional partners”. The company added that it expects “Mythos-level models will be broadly accessible within the next 6-12 months”, though it acknowledged no organization has yet developed safeguards robust enough to prevent misuse at scale.theregister