OpenAI releases GPT‑5.4‑Cyber
OpenAI unveiled a version of its flagship model tuned for defensive cybersecurity work and will restrict access to vetted security vendors, researchers and organisations. The company said access is tiered through a Trusted Access for Cyber programme with identity checks for individuals and longer onboarding for enterprises, and paid ChatGPT tiers will use smaller fallbacks when rate‑limits hit. (reuters.com, axios.com, help.openai.com)
OpenAI has released GPT‑5.4‑Cyber, a version of its flagship model tuned for defensive cybersecurity work and limited to vetted users. (openai.com) The company said on April 14 that it is expanding its Trusted Access for Cyber program to “thousands” of verified individual defenders and “hundreds” of teams that protect critical software. OpenAI said GPT‑5.4‑Cyber is the first model in that rollout. (openai.com) Access is tiered. OpenAI said individual users must complete identity checks, while enterprise customers face a longer onboarding process that includes organizational review and use-case vetting. (openai.com) Cybersecurity work here means finding and fixing weaknesses before attackers exploit them. OpenAI said the model is tuned to be more “cyber-permissive” for defensive tasks, while keeping safeguards meant to block harmful use. (openai.com) The release comes a week after Anthropic announced Mythos, its own model for security work. Reuters reported OpenAI’s launch as the latest move by major artificial intelligence labs to offer more capable tools to security professionals without opening them to the general public. (reuters.com) OpenAI has been building this access system for months. In February, it introduced Trusted Access for Cyber as a trust-based framework and said it would commit $10 million in application programming interface credits to support cyber defense work. (openai.com) The company tied the new model to a broader safety argument: stronger models can help defenders audit code, investigate alerts and respond faster, but the same capabilities can also be misused if access is too broad. OpenAI said the program is designed to place those capabilities with security vendors, researchers and organizations that can be verified. (openai.com) The launch also lands as OpenAI keeps adjusting how its mainstream products handle demand. In ChatGPT, OpenAI’s help center says paid tiers fall back to smaller models such as GPT‑5.4 mini when users hit rate limits, instead of cutting off reasoning access entirely. (help.openai.com) GPT‑5.4 itself was introduced on March 5 as OpenAI’s flagship model for professional work, with coding, tool use and a context window of up to 1 million tokens. GPT‑5.4‑Cyber takes that base system and narrows it to one job: helping approved defenders secure software before the next attack lands. (openai.com)
