OpenAI Rolls Out GPT-5.4-Cyber to Select Users Amid Tight Competition with Mythos

OpenAI has quietly introduced GPT-5.4-Cyber, a new AI model tailored for spotting software security flaws, handing early access to a handpicked bunch of cybersecurity experts. This release comes hot on the heels of Anthropic PBC's unveiling of their own cybersecurity AI tool, Mythos, marking a clear face-off in the cybersecurity AI space.

The rollout started this week through OpenAI's Trusted Access for Cyber program-launched just a couple of months ago-which offers vetted security professionals a chance to test the company's sharpest AI under fewer restrictions, spotlighting vulnerabilities in real-world systems.

In its announcement, OpenAI emphasized the model's defensive capabilities. GPT-5.4-Cyber aims to detect and help fix weaknesses before malicious actors can exploit them. The model's training focused on making it "cyber-permissive," meaning it's optimized specifically for cybersecurity tasks rather than general use.

Meanwhile, Mythos has been drawing attention not just from tech circles but also high-level financial regulators. Last week, U.S. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell reportedly issued stern warnings to Wall Street heavyweights about risks connected to Anthropic's tool. Mythos seems to have stirred both interest and unease about AI's role in critical infrastructure defense.

OpenAI pointed out a sobering truth: digital defenses have been under siege long before AI entered the picture. It noted that threat actors are now leveraging novel AI methods to probe and attack systems, underscoring the urgent need to develop equally advanced countermeasures.

Looking ahead, OpenAI plans to expand its program from the current hundreds to thousands of security specialists. The company is underscoring its dedication to broadening access while keeping a firm grip on user verification to ensure these powerful, potentially offense-capable AI tools don't fall into the wrong hands.

With competitors like Anthropic pushing forward, the pace of innovation in AI-driven cybersecurity is accelerating. What remains to be seen is how these advancements will reshape the ongoing cat-and-mouse game between defenders and attackers in the digital realm.

Meanwhile, eyes remain fixed on tech and cybersecurity firms behind these breakthroughs, as well as the broader ecosystem's response to these cutting-edge AI tools.