OpenAI expands TAC program, launches GPT-5.4-Cyber

OpenAI is scaling its Trusted Access for Cyber (TAC) program to thousands of verified defenders and hundreds of security teams while releasing a purpose-tuned variant, `GPT-5.4-Cyber`. The model is trained to be “cyber-permissive,” relaxing some refusal thresholds to support defensive tasks such as binary analysis and vulnerability research. Access is tiered and restricted: individuals verify identity at chatgpt.com/cyber and enterprises apply via account representatives. The move directly contests Anthropic’s recent Mythos rollout and shifts the competitive balance in frontier security-focused models, reducing Anthropic’s forecasted chance of being the third-best model by April 30, 2026, to **21.5%**.