Full Breakdown
EU Pushes Anthropic for Access to Super-Hacking AI Model Mythos
5/7/2026, 5:21:50 AM
Anthropic’s Mythos Model and EU Scrutiny
U.S. AI firm Anthropic unveiled Mythos in early April, a model that outperforms most humans at finding and exploiting software flaws. The European Parliament scheduled a hearing on the technology for Wednesday, but Anthropic declined, citing short-notice constraints. EU officials argue the model’s “supreme hacking abilities” require direct review to protect European digital assets.
Regulatory Context and Transatlantic Oversight Gap
The EU’s AI Office, gaining enforcement powers in August 2026, will request model access when needed, said Commission spokesperson Thomas Regnier. In the United States, deals with Microsoft, xAI, Google DeepMind and Anthropic let authorities vet advanced AI before release. The contrasting approaches underscore a transatlantic oversight gap.
Stakeholders and Access Gap
Commission tech chief Henna Virkkunen, urged by a letter from 30 MEPs including Dutch liberals Bart Groothuis and Kim van Sparrentak, must devise a mitigation plan for Europe’s “crown jewels.” ENISA seeks access to Mythos for risk assessment. Anthropic has shared the model with twelve entities, including JPMorgan Chase, but no European banks have received it.
Risks to Critical Infrastructure and Finance
EU officials warn that a super-hacking AI could threaten power grids and financial markets. Economy Commissioner Valdis Dombrovskis urged European banks to test cyber-resilience with Mythos. Eurogroup President Kyriakos Pierrakakis said AI governance needs international frameworks, a challenge amid strained multilateral relations.
Official Statements & Responses
Commission spokesperson Thomas Regnier said the AI Office will request model access when enforcement powers begin in August 2026. Henna Virkkunen, responding to the MEP letter, pledged a mitigation plan for critical sectors. Valdis Dombrovskis announced talks for banks to test cyber-resilience with Mythos. The Commission signaled “concrete action” to protect EU infrastructure from foreign actors.
Lawmakers’ Criticism
Kim van Sparrentak said Anthropic’s refusal to attend the hearing is “extremely worrying” and that “it’s absolutely vital that AI companies release models like Mythos safely.” Bart Groothuis added “Europe is not at the table,” urging stronger pressure. Markéta Gregorová warned the Cybersecurity Act could affect U.S. firms that fail to comply with EU rules.
Conflicting Reports & Gaps
Anthropic says short notice prevented its participation in the EU hearing, while EU officials claim the company has withheld model access despite repeated requests. The United States has secured voluntary review agreements with AI firms, a practice not yet mirrored in Europe, creating a transparency gap between the two jurisdictions.
Verbatim Quotes
- “extremely worrying,” — Kim van Sparrentak, Dutch MEP
- “Europe is not at the table,” — Bart Groothuis, Dutch liberal MEP
- “The challenge here is that technologies like AI necessitate international governance frameworks at a moment where multilateralism is challenged,” — Kyriakos Pierrakakis, Eurogroup President and Greece’s finance minister
- “I can see the Cybersecurity Act having an impact on U.S. companies if they don’t oblige by the rules,” — Markéta Gregorová, Czech Pirate MEP
Upcoming Steps
The EU hearing on Mythos proceeds on Wednesday, after which the Commission plans concrete actions and may require Anthropic to grant ENISA direct access for risk evaluation.
