OpenAI 或于数日内公布安全模型 GPT-6 Cyber

OpenAI will preview its next cybersecurity-focused model, GPT-6 Cyber, as soon as a few days from now, Fortune reported on Thursday. It’s also reportedly launching some kind of product that will aid customers in deploying the new model securely; that product has yet to be named, according to Fortune. The announcement is particularly notable as CEO Sam Altman has joined Anthropic CEO Dario Amodei in calling for an AI slowdown after models at Anthropic, Google, OpenAI, and Meta frontier labs all launched cyber attacks. Sources told Fortune that GPT-6 Cyber will likely preview at OpenAI’s Dev Day on Sept. 29, or possibly sooner, and launch within months. GPT-6 Cyber is already undergoing testing by one segment of OpenAI’s Daybreak preview program, in which the company gives access to its most advanced models to vetted partners for cybersecurity testing and bug-hunting. OpenAI has addressed concerns about rogue AI by announcing it will spend $1 billion subsidizing Daybreak access across the globe and as part of a “Daybreak for America” campaign, though this could be understood to be an extensive sales demo as much as it is any overall contribution to cybersecurity. It’s also worth mentioning that several of the “rogue AI” incidents in recent months have been tied to models escaping security test sandboxes, which explains why they were primed to attack something in the first place. While AI’s impact on programming remains uncertain—an insatiable hunger for AI code has overwhelmed code-review pipelines and may be lowering quality standards in the industry as a whole—security experts are bullish on AI. They’re just mixed on which side will advance faster. An EY survey released in March 2026 found 96% of corporate cybersecurity leaders believe AI is a core solution, and 95% had already deployed it, though use of AI agents was still largely exploratory. Yet 90% also believed employees who weren’t ready for AI-powered cyber attacks to be their biggest liability. A recent Institute for Security and Technology survey of national security professionals ranging from “former Under Secretaries in the U.S. government to retired flag officers” found that 72% ranked adversarial use of AI as one of their top three security concerns. Meanwhile, 78% considered cybersecurity to be one of the top three areas of promise in AI. The problem is the solution and the solution is the problem. OpenAI announced Sol and Luna, two cost efficiency-focused models, earlier this week.