Anthropic in talks with European Commission over cybersecurity models

Anthropic is in discussions with the European Commission regarding several of its artificial intelligence (AI) systems, including advanced cybersecurity models that are not yet available in the European Union.

A Commission spokesperson confirmed that the talks are part of ongoing engagement with AI developers to ensure compliance with the EU’s evolving regulatory framework. 

The company has indicated it is committed to aligning with the bloc’s general-purpose AI code of practice, which requires firms to assess and mitigate risks before deploying systems in the region.

At the center of the discussions is Anthropic’s reported development of a powerful frontier model known as “Claude Mythos.” 

The system is said to possess highly advanced cybersecurity capabilities, with internal assessments reportedly flagging it as a potential public safety risk due to its ability to autonomously identify and exploit vulnerabilities.

Instead of a broad release, Anthropic is shifting focus toward a defensive initiative, Project Glasswing, aimed at strengthening global digital infrastructure before such capabilities could be misused. 

The company is expected to limit early access of the model to a select group of users focused on cybersecurity defense, citing both safety concerns and high operational costs.

The model’s capabilities reportedly came to light following an internal configuration error that exposed unpublished materials, revealing performance levels that significantly exceed existing AI benchmarks in cybersecurity tasks.

The developments underscore the EU’s push to proactively regulate advanced AI systems. 

Under its framework, companies must evaluate and address potential risks regardless of whether the technologies are already deployed in the European market, reinforcing a precautionary approach to emerging AI threats.