Brussels is in active discussions with Anthropic regarding potential risks associated with its latest artificial intelligence model, “Claude Mythos,” amid growing concern over its advanced capabilities and possible misuse. The model has demonstrated a remarkable ability to identify software vulnerabilities, prompting the company to delay its broader release.

A spokesperson for the European Commission confirmed that regulators are seeking further details about the model’s risk profile. “A new AI system is being introduced, and it carries a range of potential risks. We require comprehensive information regarding those risks,” said Thomas Regnier. He added that initial discussions with Anthropic have already taken place, with additional meetings expected.

Earlier this month, Anthropic limited access to Mythos to approximately 40 major technology firms, aiming to provide a critical window for identifying and addressing vulnerabilities before they could be exploited. However, the exclusion of foreign entities has raised concerns about global preparedness, particularly given the model’s potential cross-border implications.

According to the company and its research partners, Mythos is capable of autonomously scanning extensive codebases to uncover and link previously unknown security flaws across a wide range of software systems, including operating systems and web browsers. Experts warn that such capabilities?executed at speeds and scales beyond human capacity?could enable large-scale cyberattacks targeting financial institutions, healthcare systems, or even national infrastructure within a matter of hours.

The announcement has drawn a mixed response from the technology community, with some analysts expressing skepticism and suggesting that heightened concern may also serve to amplify Anthropic’s competitive positioning in the rapidly evolving AI sector.

Meanwhile, senior executives from major U.S. banks reportedly convened with Jerome Powell and Scott Bessent earlier this month to assess the potential security implications of the not-yet-public model.

(The Economic Times, April 16, 2026)