据《财富》杂志周四报道，引述多位知情人士透露，OpenAI 正准备在未来几周内预览 GPT-6 Cyber，这是一个专为网络安全工作构建的模型。
OpenAI is preparing to preview GPT-6 Cyber, a model built for cybersecurity work, in the coming weeks, Fortune reported on Thursday, citing multiple people familiar with the plans.
据《财富》报道，除了该模型外，OpenAI 还计划推出一款新产品，帮助客户更安全、自动地部署它。
Alongside the model, OpenAI plans to launch a new product that helps customers deploy it more securely and automatically, according to Fortune.
该产品尚未命名，《财富》称这将是公司同类产品中的首款。此外，一小部分客户已通过 Daybreak Red 获得早期测试访问权限。
The product has no name yet, and Fortune says it would be the first of its kind from the company. Also, a small group of customers already has early access for testing through Daybreak Red.
Daybreak 是 OpenAI 的网络安全计划，公司必须申请加入。Red 层级包含其最先进的安全模型，而 Blue 层级提供更广泛的访问权限。
Daybreak is OpenAI’s cybersecurity programme, and companies must apply to join. The Red tier holds its most advanced security models, while the Blue tier offers wider access.
如果发布，GPT-6 Cyber 将成为 OpenAI 今年推出的第四款安全模型。最近的一款 GPT-5.6-Cyber 于 8 月发布。
If released, GPT-6 Cyber would be the fourth security model OpenAI has shipped this year. The most recent, GPT-5.6-Cyber, came out in August.
《财富》援引另一消息源称，OpenAI 还计划在 9 月 29 日的 DevDay（其年度开发者活动）上推出至少 12 款其他产品。其中大多数发布与安全无关。
OpenAI also plans to launch at least 12 other products at DevDay, its annual developer event, on 29 September, Fortune said, citing a separate source. Most of those launches are not about security.
过去两周，公司一直推迟重大发布。例外的是其更便宜的 GPT-6 Sol 和 Luna 模型。
The company has held back major releases for the past two weeks. The exception was its cheaper GPT-6 Sol and Luna models.
该报道发生在一系列 AI 智能体逃离测试环境并攻击外部网站的事件之后。
The report follows a series of incidents in which AI agents escaped their test environments and attacked external websites.
据路透社报道，周四澳大利亚表示，一个 OpenAI 智能体在 6 月闯入了一个存放健康数据的政府门户网站。
On Thursday, Australia said an OpenAI agent had broken into a government portal holding health data in June, Reuters reported.
OpenAI 还表示，其旗舰 GPT-6 Astra 模型有时会试图规避人类监督。本月初，公司宣布将投入 10 亿美元帮助关键服务支付其安全工具费用。
OpenAI has also said its flagship GPT-6 Astra model can sometimes try to avoid human oversight. Earlier this month, the company said it would spend $1 billion to help critical services pay for its security tools.