OpenAI launches cyber model
OpenAI released a version of its flagship model tuned specifically for defensive cybersecurity work and is limiting early access to vetted security vendors and researchers. The rollout is part of a controlled programme called Trusted Access for Cyber and OpenAI says access will expand gradually to organisations that pass verification processes. (reuters.com)
OpenAI has launched GPT-5.4-Cyber, a version of its flagship model tuned for defensive cybersecurity work and kept off general public release. (openai.com) The model went live on April 14, 2026 through OpenAI’s Trusted Access for Cyber program, which now covers thousands of verified individual defenders and hundreds of teams protecting critical software. OpenAI said higher verification tiers unlock more capable cyber features. (openai.com) Reuters reported the first users will be vetted security vendors, organizations and researchers because GPT-5.4-Cyber is designed to be more permissive on sensitive tasks such as vulnerability research and analysis. Users approved for the highest tier get access to the new model. (reuters.com) Cybersecurity work here means finding weak spots in software before criminals do. OpenAI said it is fine-tuning models to help defenders analyze code, investigate vulnerabilities and secure systems as more capable models arrive over the next few months. (openai.com) The guarded rollout follows a broader shift in artificial intelligence security: companies are building models that can help with cyber defense but could also be misused for intrusion or malware work if released too widely. OpenAI said Trusted Access is meant to place stronger capabilities “in the right hands” through identity and trust checks. (openai.com) The timing is not accidental. Reuters said OpenAI announced GPT-5.4-Cyber about a week after Anthropic introduced Mythos, its own frontier model for cybersecurity, with access limited over fears hackers could abuse it. (reuters.com, cnbc.com) OpenAI launched Trusted Access for Cyber in February 2026 and tied it to a $10 million pool of application programming interface credits for security work. The company said the program was built to expand defensive use while reducing misuse risk through verification and monitoring. (openai.com) OpenAI said access will widen gradually as more organizations complete the verification process. For now, the company is treating GPT-5.4-Cyber less like a mass-market chatbot and more like restricted security equipment. (openai.com, reuters.com)
