The White House has stepped directly into OpenAI's product launch calendar.根据 6 月 25 日发布的多份报告，特朗普政府要求 OpenAI 推迟并错开其下一个人工智能模型 GPT-5.6 的发布，理由是官员们希望在该技术普及到广泛受众之前有时间评估网络安全风险。

The request came on June 25 and follows an executive order President Donald Trump signed on June 2 that established a voluntary framework allowing government cybersecurity teams up to 30 days to assess advanced AI models before they go live for partners and the public. The move signals a new, more hands-on posture from Washington toward the release cadence of the country's most prominent AI lab. 更多相关背景，请参阅我们的AI新闻。

该报道得到了 Axios、路透社、TechCrunch、彭博社、The Information 和 BeInCrypto 等主要媒体的证实，所有媒体都援引了熟悉讨论情况的人士的话。

新的审核流程是什么样的

关于 OpenAI 的惊人发布的讨论涉及两个重要办公室的官员：国家网络主任办公室和科学技术政策办公室。 Their concern, broadly, is that a frontier model with GPT-5.6's anticipated capabilities could be exploited by adversarial nations or non-state actors if deployed widely without adequate security vetting.

据报道，OpenAI 首席执行官萨姆·奥尔特曼 (Sam Altman) 回应称，其发布将转向有限预览，仅向选定的合作伙伴群体开放，而不是立即全面推出。 The staggered approach would give the government's review window room to run while still allowing OpenAI to begin controlled distribution.

OpenAI is not the only company receiving this kind of attention. Anthropic, the AI safety-focused competitor behind the Claude model family, has reportedly faced restrictions on its own advanced models amid similar national security evaluations, according to the same reports.

请求背后的行政命令

The June 2 executive order was designed with a specific geopolitical backdrop in mind.过去几年，中美技术竞争愈演愈烈，华盛顿的担忧很直接：具有强大功能的先进人工智能模型，如果在没有足够保障措施的情况下发布，可能会被武器化或被盗。

The order's voluntary review process represents a deliberate middle path.它避免了彻底禁止或强制发布前政府批准的严厉做法，这种立场曾令部分行业感到震惊。 An earlier draft of the executive order, which proposed more stringent measures, was postponed in May 2026 amid significant pushback from the AI sector.

Instead, the finalized framework invites companies to submit their most powerful models for a government assessment before public launch. The initiative also appears designed to foster industry partnerships that could eventually translate into government contracts, as federal agencies integrate AI tools into defense, intelligence, and civilian operations.

发布时间表中的罕见干预

For OpenAI, the request lands at a delicate moment. The company has been preparing what it has described as one of its most capable models to date, with GPT-5.6 expected to extend the context window and coding abilities that distinguished the GPT-5 generation. Any delay, even a partial one, reshapes a competitive landscape in which Google, Anthropic, and a fast-moving cohort of Chinese labs are all racing to release frontier systems.

转向有限预览而不是全面亮相的决定值得注意，因为它将产品发布重新定义为私营公司和联邦政府之间的协调行动。 That is a relatively new dynamic in the AI industry, where releases have historically been timed to competitive pressure and developer demand rather than a security review clock.

这对人工智能领域意味着什么

For the broader AI industry, a voluntary 30-day review window is far less disruptive than the mandatory licensing or pre-market approval regimes that some members of Congress have floated.投资者和高管普遍欢迎这种较为温和的做法，认为这是在不限制创新的情况下解决国家安全问题的一种方式。

同时，该事件开创了先例。如果政府能够有效地制定世界上最有价值的人工智能公司 OpenAI 的发布时间表，那么随着模型能力的增强，较小的实验室可能会面临类似的期望。自愿合作与事实上的监管之间的界限在实践中可能比表面上看起来更细。

更大的问题是，审查过程是否发现了真正的风险，或者只是减缓了该领域的发展。政府将 30 天的窗口期设定为防止滥用的保障措施。批评者会辩称，无论华盛顿与美国公司进行什么谈判，坚定的对手都可以从国外获得强大的开放权重模型。安全与速度之间的紧张关系不太可能通过单一行政命令得到解决。

目前，GPT-5.6 似乎即将首次亮相。 OpenAI 尚未对具体时间公开发表评论，但向交错发布的转变表明，模型意外发布的时代可能正在让位于更加深思熟虑且越来越受到更多监督的时代。

---

掌握AI最新动态

获取最新的AI新闻、分析和突破——尽在一处。

阅读更多AI新闻 →