OpenAI has launched GPT-5.4-Cyber, a variant of its GPT-5.4 model tailored for defensive cybersecurity, accessible exclusively to verified security professionals under its “Trusted Access for Cyber” (TAC) program. This enhanced model allows for activities like binary reverse engineering while implementing stricter access controls, particularly for third-party platforms. In tandem with the release, OpenAI promotes its existing Codex Security tool, which has effectively identified and patched over 3,000 critical vulnerabilities since its inception. Additionally, OpenAI has committed $10 million to its Cybersecurity Grant Program, benefiting over 1,000 open-source projects with complimentary security scanning services. The unveiling of GPT-5.4-Cyber follows competitor Anthropic’s launch of its own AI model, Claude Mythos, focused on exploiting system vulnerabilities, raising alarms among financial institutions and U.S. government entities. The cybersecurity landscape remains vigilant as these innovations evolve.
Source link
