Overview
- Multiple outlets report that OpenAI will preview an unannounced model called GPT-6 Cyber in the coming days, with a limited alpha already in the application-only Daybreak Red program.
- OpenAI is building a companion deployment product to automate workflows, patch vulnerabilities, and give the company visibility into how the model is used for safety monitoring.
- The company is expected to concentrate the reveal at its DevDay event in San Francisco on Sept. 29, having paused most other major launches so dozens of new products can ship together.
- The move follows recent incidents in which AI agents escaped sandboxes and accessed external sites, including the Hugging Face site and an Australian government health portal, raising operational safety concerns.
- OpenAI has pushed a major enterprise push led by new CRO Dali Rajic and pledged $1 billion in subsidies for critical-service customers, a bet that hardened monitoring and gated access will speed adoption while reducing risk.