Solutions  ·  2026-04-30

OpenAI Expands AI-Assisted Cyber Defense Access to All Vetted Government Levels

SolutionsMedium impactGlobal
OpenAI announced expansion of its Trusted Access for Cyber program from a select group of partners to all vetted government levels (federal, state, and local), providing approved agencies with access to special versions of its models with reduced guardrails for defensive cybersecurity purposes. The company published a proposed action plan titled 'Cybersecurity in the Intelligence Age' and held hands-on workshops with federal agencies including the Pentagon, White House, DHS, and DARPA to test the model's capabilities. OpenAI also met with the White House National Cyber Director alongside Anthropic and other tech firms to discuss AI and cybersecurity.
This represents a sharp policy contrast to Anthropic's tightly-controlled Mythos release via Project Glasswing. OpenAI is advocating for broad democratization of offensive-capable AI models for defensive use, arguing that restricting access to Fortune 50 firms leaves under-resourced agencies vulnerable. The approach mirrors wider AI industry debates between rapid innovation versus cautious deployment. For security practitioners, the announcement signals that frontier-model cyber capabilities will become available at scale to vetted defenders within months, not years.
State and local government cybersecurity teams eligible for vetting should assess whether their current vulnerability management workflows can integrate AI-assisted discovery tools. Consulting practices advising public-sector clients need to prepare guidance on responsible use of reduced-guardrail models, including human-in-the-loop requirements and disclosure policies. Private-sector organizations should monitor whether OpenAI's Trusted Access expands beyond government to regulated critical infrastructure sectors.
Sources
CNN — OpenAI expands cyber model access to governmentAxios — OpenAI and Anthropic brief House Homeland Security
See this in the live feed Explore related AI security and governance findings — updated every morning.
Open the feed →