OpenAI Expands Its Models to Europe, Anthropic Holds Firm

OpenAI Opens Up to Europe, Anthropic Remains Cautious

OpenAI has recently made a significant decision by agreeing to partially open its artificial intelligence models to European institutions, a move that major labs had been reluctant to make until recently. This opening specifically concerns GPT-5.5-Cyber, a specialized version of its latest model, which will be accessible to select European partners, including governments, cyber authorities, businesses, and community institutions such as the EU AI Office. However, this access will be limited to cybersecurity teams validated by OpenAI.

In contrast, Anthropic has chosen to keep its Mythos model out of reach for European institutions. This advanced cyber model, launched a month ago, has raised concerns about its potential to automate attacks against critical infrastructure. Despite Brussels' attempts to gain early access to assess its capabilities and security mechanisms, Anthropic remains firm in its decision not to share its model.

A New Balance of Power at Play

This divergence in approach between OpenAI and Anthropic highlights a new balance of power between American labs and European authorities. The European Commission has expressed that discussions with Anthropic are not as advanced as those with OpenAI, thus emphasizing a desire to intervene early in the development of the most advanced AI models.

The goal of the EU AI Office is to create a political and regulatory precedent by obtaining institutional access to systemic models to understand their capabilities, limitations, and security mechanisms. Brussels also wants visibility on red teaming procedures, the safeguards put in place by labs, and post-deployment monitoring mechanisms.

Towards Enhanced Oversight

The European Commission is seeking to establish oversight of advanced artificial intelligence models, similar to the controls applied in critical sectors like nuclear energy or finance. Cybersecurity is the first area of experimentation, as models like GPT-5.5-Cyber or Mythos are perceived as capacity multipliers, capable of accelerating the discovery of vulnerabilities and the generation of exploits.

OpenAI seems to have anticipated this political evolution by strengthening its institutional ties with European states. In Europe, access to the market for advanced models will depend as much on regulatory cooperation as on technological quality. Labs that work closely with Brussels could gain a significant advantage with administrations and critical operators.

Anthropic, on its part, is taking a cautious approach, avoiding excessive exposure of its actual capabilities and the creation of regulatory precedents. This caution raises a central question for all American labs: how far to accept institutional scrutiny without compromising their technological edge?

A Shift in Doctrine

This situation marks a shift in doctrine in the relations between Europe and American tech. Brussels no longer wants to regulate after the fact but to intervene proactively in the face of critical risks. Cybersecurity may just be the beginning, with similar debates to come in areas such as scientific, biological, financial, or military models. A new balance of power is emerging between states and private actors for the control of critical infrastructures in the age of artificial intelligence.