OpenAI has introduced GPT-6 Cyber, a model focused on cybersecurity tasks. According to the announcement, the model is built to locate exploitable vulnerabilities and to develop fixes for them. OpenAI states that the model was trained to operate within authorized scope, framing it as an effort toward aligned cybersecurity agents.
Why it matters
Security work involves both offensive discovery and defensive remediation, and constraining a model to an authorized scope addresses a core concern about tools capable of identifying exploitable weaknesses. The emphasis on alignment reflects the dual-use nature of vulnerability research.
Who should care
Security teams, defenders, and organizations evaluating AI-assisted vulnerability detection and remediation may find this announcement relevant.