EU Commission Demands Direct Access to OpenAI and Anthropic AI Models for Regulatory Review

The European Commission presses OpenAI and Anthropic for direct AI model access. OpenAI offers GPT-5.5-Cyber access to EU institutions and cyber authorities. Anthropic holds its first Commission meeting on May 7, 2026, but withholds preview access to Mythos. Parliament officials discuss Mythos dangers with Commission officials on May 13, 2026. EU AI Office enforcement powers begin in August 2026.

OpenAI model access offer

• ▪OpenAI has offered the European Commission access to GPT-5.5-Cyber for regulatory review.
• ▪OpenAI's GPT-5.5-Cyber access offer extends to EU institutions, European governments, European businesses, and European cyber authorities.
• ▪The European Commission has welcomed OpenAI's cooperation on model access.

Anthropic negotiation status

• ▪The European Commission held a first meeting with Anthropic on May 7, 2026.
• ▪Anthropic's talks with the European Commission have not yet reached system-access negotiations.
• ▪The European Commission has not been granted preview access to Anthropic's Mythos model.
• ▪Thomas Regnier stated that the European Commission is reaching out to Anthropic and has received certain information.

Mythos security concerns

• ▪European Parliament officials will discuss the dangers of the Mythos model with European Commission officials and ENISA on May 13, 2026.
• ▪Anthropic limited Claude Mythos access to 40 major tech players in April 2026 so vulnerabilities could be fixed before attackers exploited them.
• ▪Thomas Regnier stated that Anthropic's new AI model comes with a certain number of risks and the European Commission needs information about these risks.

EU regulatory enforcement timeline

• ▪The European Commission is pressing OpenAI and Anthropic for direct access to advanced AI models for regulatory review.
• ▪The European Commission wants access that lets officials test safety claims against real deployments rather than company summaries alone.
• ▪The European Union published its first general-purpose AI code draft in 2024.
• ▪The EU AI Office's enforcement powers start in August 2026.

Perspective of European Commission

• ▪The European Commission wants access that lets officials test safety claims against real deployments rather than company summaries alone.
• ▪Thomas Regnier stated that Anthropic's new AI model comes with a certain number of risks and the European Commission needs information about these risks.

2 sources