Original article appeared here.
Anxious Security Pros Watch as Anthropic, OpenAI Expand Access to Frontier AI Models
Avatar photoby Jeffrey Burt on June 3, 2026
Frontier AI models from Anthropic and OpenAI continue to send ripples through the cybersecurity world, with both AI vendors expanding access across to organizations across the globe while the White House looks to exert more oversight of them.
The moves highlight the anxiety generated by Anthropic’s unveiling in April of its Claude Mythos Preview model, which demonstrated extraordinary capabilities in not only detecting and identifying software vulnerabilities but also to quickly develop exploits for them.
In the wrong hands, Mythos – which found tens of thousands of security flaws in the first few weeks after its introduction – could be a dangerous tool for threat actors, which led Anthropic to create Project Glasswing and limit access to several dozen security vendors and researchers to use to develop advanced cybersecurity tools. OpenAI took a similar path, most recently with GPT-5.5-Cyber.
Anthropic executives have slowly been increasing who has access to Mythos, and this week said they were expanding Glasswing membership from the original 50 or so to about 150 organizations based in more than 15 countries as well as ENISA, the European Union’s cybersecurity agency.
Meanwhile, OpenAI reportedly is offering major banks in the U.K. access to GPT-5.5-Cyber after Anthropic blocked them from using Mythos. The AI Security Institute in late April said testing of both Mythos and GPT-5.5-Cyber found that both reach similar levels of performance.
Bonfy CEO Gidi Cohen said the executive order “reflects something the security community has understood for a while: frontier AI models are no longer just productivity tools. They are infrastructure with national security implications.”
Read the rest of the article here.