OpenAI's GPT-5.4-Cyber Rejects Mythos Playbook

OpenAI has introduced GPT-5.4-Cyber, a more permissive version of its flagship model designed specifically for defensive cybersecurity work. The release comes in direct contrast to Anthropic’s recent Mythos rollout, which is restricted to a small group of approved organizations.

OpenAI cyber researcher Fouad Matin framed the company’s position clearly, saying that cybersecurity is a “team sport” and that

“no one should be in the business of picking winners and losers”

when it comes to who gets to defend their systems.

A Broader Access Strategy

GPT-5.4-Cyber is being rolled out to thousands of verified defenders through OpenAI’s Trusted Access for Cyber initiative. Anyone who passes identity checks can gain access, marking a significantly wider distribution than Anthropic’s Mythos.

• OpenAI is opening access to verified users who complete ID checks via its Trusted Access for Cyber program.
• Anthropic’s Mythos model is limited to a whitelist of just over 40 trusted partner organizations.
• The contrasting approaches highlight a philosophical divide over how advanced cyber-capable AI systems should be distributed.

What GPT-5.4-Cyber Can Do

The new model is purpose-built for defensive security tasks. One of its key capabilities is the ability to reverse-engineer compiled software to identify malware and security vulnerabilities. This allows analysts to inspect and assess programs even when the original source code is unavailable.

By focusing on defensive applications, OpenAI is positioning GPT-5.4-Cyber as a tool to strengthen security teams rather than concentrate capabilities within a small circle of large institutions.

Rising Stakes in AI Cybersecurity

The broader debate over access comes amid heightened scrutiny of advanced AI systems with cyber capabilities. Treasury Secretary Bessent recently convened Wall Street leaders for an emergency briefing on Anthropic’s Mythos, as concerns grow about the hacking potential of cutting-edge AI models.

It remains unclear how GPT-5.4-Cyber will compare to Mythos on benchmark performance. However, the divergence in rollout strategies signals that competition between major AI labs is no longer just about model performance — it’s also about governance, access, and the role AI will play in national and corporate cybersecurity.