OpenAI expands GPT-5.4-Cyber access to vetted users
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI released GPT-5.4-Cyber, a new model variant designed to assist with defensive cybersecurity tasks and reduce friction on dual-use queries for vetted users.
- OpenAI is expanding its Trusted Access for Cyber program, adding verification tiers that unlock more powerful capabilities, including vulnerability research, for approved individuals and security teams.
- Anthropic has taken a more restrictive approach by limiting access to its Mythos Preview model to about 40 organizations due to concerns over its exploit-finding capabilities.
- Fouad Matin, a cyber researcher at OpenAI, stated the company believes cybersecurity defense should be widely empowered, not limited to select 'winners.'
- OpenAI is not currently offering GPT-5.4-Cyber access to U.S. government agencies but says it is in discussions and will evaluate access through internal safety reviews.
- Security experts note that while AI-identified vulnerabilities are often not novel or easily exploitable, the speed and advancement of these models are raising concerns among government and business leaders.
Why it matters: OpenAI’s push to broaden access to high-risk AI cyber tools—unlike Anthropic’s tight restrictions—means more defenders can use advanced AI, but also increases the stakes if verification fails. The model’s compute demands and access delays may limit immediate adoption, even as scrutiny from officials grows.
