OpenAI is finalizing a cybersecurity-focused model for a limited release to
select companies, similar to Anthropic’s approach with its Mythos Preview. AI
companies are cautious about releasing models with advanced capabilities due to
potential misuse, as these tools can autonomously identify flaws or develop
exploits in code. Experts warn that despite restrictions, such capabilities will
soon be publicly accessible. Security firms compare this cautious rollout to
responsible vulnerability disclosure in cybersecurity, though it’s unclear if
OpenAI will later expand access to its model.
