OpenAI unveiled its GPT-5.4-Cyber model to about 50 cyber defense practitioners from federal agencies, state governments, and Five Eyes allies in Washington, D.C., under a tiered access approach previously announced.

The briefing signals rising government and enterprise interest in AI-powered cybersecurity, with competitors like Anthropic pursuing parallel initiatives.

Anthropic has launched Project Glasswing, allowing partners to preview its unreleased model as part of the broader race to field AI cybersecurity tools.

Officials say the tool can help identify vulnerabilities in outdated government information systems, highlighting a focus on modernizing security postures.

Industry observers note that governments and companies want AI to speed up discovery of security flaws while balancing risk around legacy systems and deployment safeguards.

OpenAI has not provided public comment on the briefing, and Reuters was unable to independently verify all details or obtain confirmation from OpenAI.

The development raises questions about how governments will integrate advanced AI defenses into cyber strategies against increasingly sophisticated threats.

Details on the specific functions or policy implications of the cyber product were not disclosed in the available briefings.

Industry context includes findings that AI-driven attack capabilities improve with more compute, better orchestration, and tighter integration with external tools.

There are reports that the NSA may test Anthropic’s Mythos Preview for cyber defense despite Pentagon classifications labeling the technology as a supply-chain risk.

Anthropic’s Mythos Preview has a restricted rollout due to cyber risks, with NSA reportedly testing Mythos despite concerns.

The cyber-AI race unfolds amid concerns about misuse against critical infrastructure, prompting ongoing government engagement to ensure secure and sensitive deployments.