OpenAI rolls out GPT‑5.4‑Cyber

OpenAI has released GPT‑5.4‑Cyber, a cybersecurity version of its GPT‑5.4 model, and widened access to “thousands” of vetted defenders. (openai.com) OpenAI said on April 14 that the model is tuned for defensive work such as secure code review, malware analysis and reverse engineering, and that it will not be offered as a general ChatGPT product. Reuters reported the launch the same day and said it followed Anthropic’s recent Mythos announcement. (openai.com) (money.usnews.com) Cybersecurity work here means using an artificial intelligence model as a software assistant for defenders: reading code, spotting weak points, and helping analysts understand how an attack might work before an attacker uses it. OpenAI said GPT‑5.4‑Cyber has already helped security teams identify and fix more than 3,000 vulnerabilities. (openai.com) (thehackernews.com) OpenAI tied the release to its Trusted Access for Cyber program, which screens users before granting broader model permissions for high-risk security tasks. The company said the expansion moves that program from a smaller pilot to access for thousands of verified professionals. (openai.com) The timing follows a week in which Anthropic’s Mythos model pushed the industry conversation toward what advanced models could do in software exploitation and defense. Wired reported that OpenAI said its safeguards “sufficiently reduce cyber risk” for this release, even as Anthropic kept Mythos tightly restricted. (wired.com) (money.usnews.com) That shift is showing up outside the model labs. Stocktwits, citing reporting on April 15, said Coinbase shares rose more than 4% after Chief Security Officer Philip Martin said the exchange was in talks with Anthropic, while Coinbase and Binance were described as preparing for stronger artificial intelligence-driven cyber threats. (stocktwits.com) OpenAI has been building the policy scaffolding for this kind of release for months. Its updated Preparedness Framework says frontier models are evaluated for severe-risk areas including cybersecurity, and models that cross higher risk thresholds face deployment blocks or stronger controls. (openai.com) GPT‑5.4 itself arrived on March 5 with a one million token context window and stronger coding and computer-use features, which are the base capabilities that make a cyber-focused variant more useful to defenders. OpenAI’s model documentation says GPT‑5.4 and GPT‑5.4 Pro support about 1.05 million tokens of context. (openai.com) (developers.openai.com) OpenAI’s pitch is that broader defender access can help close security gaps faster than attackers can exploit them. The company is betting that screening, logging and use restrictions are enough to let more security teams use a stronger model without turning it into a public offensive tool. (openai.com)