OpenAI 证实，应川普政府白宫的要求，该公司已推迟其下一代 AI 模型 GPT-5.6 的公开发布。OpenAI 将首先把该模型分享给由美国政府预先批准的少数客户，并对此延期发布表示无奈，但认为该限制只是暂时的，并希望在未来几周内向大众开放。

此前川普总统签署了一项旨在解决强大 AI 模型网络安全担忧的行政命令，要求 AI 实验室在广泛发布前 30 天与政府分享模型。然而，OpenAI 表示目前尚未建立此类自愿性框架，这使得 frontier AI 实验室处于尴尬的过渡期。此外，白宫先前也对 Anthropic 发出了出口管制指令，迫使其下架最先进的模型。

据悉，GPT-5.6 将推出 Sol、Terra 和 Luna 三种版本，其中 Sol 是在网络安全、生物学和代理能力基准测试中最强大的版本。OpenAI 强调该模型配备了分层安全防御栈，以防止恶意行为者将其用于网络攻击等不当行为。

OpenAI confirmed it has delayed the public release of its next-generation AI model, GPT-5.6, at the request of the Trump administration's White House. The company will initially share the model with a group of customers preapproved by the US government, viewing this delay as temporary and hoping for a broader release in the coming weeks.

This delay follows an executive order signed by President Trump targeting cybersecurity concerns of powerful AI models, requiring labs to share models 30 days prior to public release. However, OpenAI noted that no such voluntary framework actually exists yet, leaving labs in an awkward interim, especially after the White House recently forced Anthropic to take its advanced models offline.

GPT-5.6 will be released in three versions: Sol, Terra, and Luna, with Sol being the most capable version in benchmarks testing cybersecurity, biology, and agentic abilities. OpenAI highlighted that the models feature a layered safeguard stack designed to prevent bad actors from utilizing them for malicious activities like cyberattacks.