EU Seeks U.S. Talks Over AI Safety as Anthropic Plans Mythos Rollout

TLDR

• The EU wants deeper talks with U.S. officials on advanced AI models with cyber capabilities.
• Anthropic has not granted the EU AI office preview access to its Mythos model.
• U.S. officials are weighing model access, innovation, cyber safety, and competition with China.
• Anthropic says Mythos-class models need strong cyber safeguards before wider release.
• Anthropic expects to bring Mythos-class models to customers in the coming weeks.

The European Union plans to expand talks with the U.S. administration over advanced AI models with cyber capabilities. The discussions center on access, safety reviews, and rules for powerful systems such as Anthropic’s Mythos model. The talks come as Anthropic prepares to bring Mythos-class models to customers in the coming weeks.

EU Pushes for Access to Advanced AI Reviews

According to CNBC, a European Commission official said the bloc wants stronger discussions with U.S. officials on frontier AI models. The talks cover systems that can support complex cyber tasks and security testing. The EU has not received preview access to Anthropic’s Mythos model. Its AI office also has not reviewed the model under the current access process.

Anthropic gave early preview access to selected companies and organizations through Project Glasswing. However, government access outside the U.S. remains limited, except for the U.K. AI Security Institute. A person familiar with the talks said Anthropic told EU officials to seek U.S. permission. The person said the White House did not object to the EU alone.

The same person said U.S. officials had wider concerns about sharing Mythos with non-U.S. governments. The talks now place model access inside a broader transatlantic AI security discussion.

U.S. Weighs Innovation and Safety Concerns

The White House has worked with AI labs on advanced model deployment. U.S. officials want to maintain the country’s lead while managing security risks. Treasury Secretary Scott Bessent said the administration works “very closely” with AI labs. He also said officials want to balance innovation and safety.

The administration opposed a wider expansion of Mythos access, according to earlier reports. Anthropic had proposed giving roughly 70 more companies and organizations access to the model.

Mythos drew attention after Anthropic announced it in April. The model raised concerns because of its reported ability to identify software vulnerabilities at scale. Anthropic CEO Dario Amodei warned this month about a narrow security window. He said organizations had six to 12 months to patch many vulnerabilities before Chinese AI catches up.

Anthropic Plans Wider Mythos-Class Rollout

Anthropic said Thursday that models with Mythos-level capability need strong cyber safeguards before broad release. The company also said it has made fast progress on those protections. “We’re making swift progress on developing these safeguards,” Anthropic said. The company expects to bring Mythos-class models to all customers in the coming weeks.

The European Commission has also expanded technical talks with Anthropic and other model developers. These talks involve companies that notified their latest models to the EU AI office. Commission spokesperson Thomas Regnier said cybersecurity remains a shared priority. He said the EU and U.S. agreed to recognize their respective standards in this area.

The discussions now cover model access, cyber safeguards, and review procedures. They also come as governments examine how advanced AI systems handle security-sensitive capabilities. Anthropic’s next rollout will test how labs manage access to powerful models. The EU continues technical discussions with Anthropic while seeking closer talks with U.S. officials.

The post EU Seeks U.S. Talks Over AI Safety as Anthropic Plans Mythos Rollout appeared first on Blockonomi.