OpenAI just deployed GPT-5.4-Cyber, a specialized model aimed at autonomously discovering software bugs, to a vetted group of customers. This move arrives only one week after rival Anthropic released its similar Mythos tool, intensifying the race among AI labs to build defensive capabilities against emerging cyber threats.

Concerns are mounting that powerful generative models could assist malicious actors in exploiting vulnerabilities, potentially rendering current security measures obsolete. Anthropic stated its model already flagged thousands of severe flaws across major operating systems and browsers. This arms race worries regulators, prompting Treasury Secretary Scott Bessent and Fed Chair Jay Powell to convene bank leaders over the risks.

OpenAI limited initial access to its highest-tier trusted program members, emphasizing the need to prevent misuse while expanding availability later. The company previously reported its Codex Security product helped resolve over 3,000 high-priority vulnerabilities earlier this year. Investors recall the market jitters that followed Anthropic’s announcement, spooked by the technology’s sector implications.

Both firms frame these tools as essential countermeasures against hackers increasingly leveraging AI for offensive operations. The controlled release strategy attempts to balance capability enhancement with responsible deployment in a highly sensitive area of digital defense.