Overview
- The Five Eyes intelligence and cyber agencies issued a joint advisory on June 22 warning that frontier AI models will transform offensive and defensive cyber capabilities on a timeline of months rather than years.
- In response to national security directives, Anthropic has suspended or restricted access to its highest‑capability models, including Mythos 5 and Fable 5, to comply with U.S. controls.
- OpenAI has released cybersecurity‑focused models (GPT‑5.4‑Cyber and GPT‑5.5‑Cyber) and expanded vetted access programs without the same foreign‑access restrictions applied to Anthropic.
- Independent tests and security firm reports show these frontier models can find and chain software vulnerabilities far faster than human teams, with Palo Alto reporting weeks of work that would have taken a year and the UK AI Security Institute finding high exploit rates in trials.
- The Five Eyes advisory urges concrete action—use AI defensively, speed patching, remove unnecessary internet exposure and tighten access controls—while raising urgent questions about governance, model diffusion and support for smaller organisations that lack strong cyber defences.