The EU is testing Anthropic's most powerful cyber model, but did not reach its newest version

The EU cybersecurity agency ENISA has been granted access to Anthropic's Mythos 5 model and has begun testing it. This was announced on Thursday 10 September 2026 by European Commission spokesperson for digital affairs Thomas Regnier, who said that following the Commission's constructive engagement with Anthropic it could confirm that ENISA had been granted access and was now testing the model. According to the spokesperson, ENISA has also been granted access to OpenAI's latest model, GPT-6-Astra. Mythos 5, previewed by Anthropic in April 2026, can identify and exploit vulnerabilities with unusual speed, which is why the company admits only selected partners to it through Project Glasswing, run in cooperation with the US government; ENISA is the first European organisation admitted to the programme. The talks took about three months and were complicated by US export controls on advanced AI technologies. According to further reporting, the access did not extend to the newest version of the model, Mythos 5.1, released at the beginning of September.