16, April 2026
OpenAI is releasing its latest cybersecurity model, GPT-5.4-Cyber, to a limited number of partners, following a similar move by rival Anthropic.
The restricted release reflects concerns about AI-enabled cyber threats.
*Key Developments:*
– OpenAI’s GPT-5.4-Cyber will be available to “the highest tiers” of people and organisations in its Trusted Access for Cyber (TAC) scheme.
The model is “trained to be cyber-permissive”, allowing defenders to test their systems for vulnerabilities with fewer refusals.
Anthropic’s Claude Mythos model was released to 40 major tech players, uncovering thousands of vulnerabilities in widely-used software.
Some vulnerabilities had gone unnoticed for years or decades, highlighting the model’s capabilities.
*Industry Concerns:*
Major American bank chiefs met US Treasury Secretary Scott Bessent and Federal Reserve Chairman Jerome Powell to discuss the system’s dangers to the financial sector.
The releases come amid excitement in Silicon Valley about generative AI’s growing capability in producing and evaluating computer code.
Companies’ Stances:
OpenAI: “Our goal is to make these tools as widely available as possible while preventing misuse.”
Anthropic: Strict access limits give defenders a head start in fixing vulnerabilities before attackers can exploit them.
OpenAI: “We don’t think it’s practical or appropriate to centrally decide who gets to defend themselves. Instead, we aim to enable as many legitimate defenders as possible.”
Access and Safety:
GPT-5.4-Cyber will be available to thousands of verified individual defenders and hundreds of teams responsible for defending critical software.
Developers add safeguards to prevent malicious use, but concerns remain about AI-enabled cyber threats.
