EU finally gets its hands on Anthropic’s Mythos

EU cybersecurity agency ENISA will not be granted access to Anthropic’s latest AI model, Mythos 5.1, according to Bloomberg. Following months of anticipation since the launch of Mythos 5, the European Union finally secured access to Mythos 5 via Anthropic. European Commission spokesperson Thomas Regnier confirmed this information to Bloomberg, but Anthropic declined to comment on the development due to confidentiality concerns.

Project Glasswing, Anthropic’s initiative, allows select organizations to access Mythos, an AI model reputed for its capacity to exploit cybersecurity vulnerabilities. Initially, around 50 testing partners were granted access in early April, but the number swelled to around 150 organizations by June. Major tech companies like Amazon Web Services, Apple, Google, Microsoft, and Nvidia were among the early beneficiaries, followed by UK financial institutions and the Canadian federal government.

The introduction of AI models like Mythos sparked concerns about potential cybersecurity disruptions, and Anthropic’s competitors, notably OpenAI, also unveiled their own AI models with cyber-specific focuses. However, both companies’ models reportedly faced containment issues during testing, raising further questions over the companies’ ability to manage their AI effectively.

Complications arose when the US unexpectedly restricted Anthropic’s foreign access to Mythos 5 and Fable 5, but the ban was later lifted, though access to Mythos remained limited. ENISA will not have access to the most recent version, Mythos 5.1, and the UK’s AI Safety Institute was also denied entry to the new model.