在 2026 年 3 月 4 日（GMT+8）报导的内部全员会议上，Sam Altman 对 OpenAI 员工表示，公司对美国 Defense Department 的 AI 软体作战运用没有决策权。依据知情人士说法，Defense Department 明确表示会听取 OpenAI 的技术应用专业意见，但不希望公司对特定军事行动是否为好或坏主意发表立场。Altman 并暗示，这种决策边界可能是 Pentagon 与 Anthropic 之间紧张关系的一部分。

这场周二会议是 OpenAI 在上周五深夜达成协议后，Altman 首次回应员工提问；该协议允许 Pentagon 在其机密网路部署 OpenAI 模型，时间间隔约 4 天。此前 Pentagon 与 Anthropic 曾出现对峙，Anthropic 要求其技术不得用于对美国人的大规模监控，且不得用于完全自主武器的部署。报导亦提到 Anthropic 曾就委内瑞拉总统 Nicolas Maduro 抓捕突袭中的技术使用提出问题，但 Anthropic 否认曾与 Defense Department 讨论特定行动。

Altman 早先称，与 Defense Department 的安排符合 OpenAI 原则，即禁止国内美国人大规模监控，并要求在武力使用上维持 human responsibility（含 autonomous weapon systems）。其后他又称此交易看来 opportunistic and sloppy，并表示正与部门增补条款，以更清楚界定原则，包括确保 AI 不被用于国内监控，且 National Security Agency 等情报机构不能依赖 OpenAI 服务。Altman 也表示仍在推动 Pentagon 移除对 Anthropic 的 supply-chain risk 标记；该标记被描述为此前未曾给予任何美国公司。

In an internal all-hands reported on March 4, 2026 (GMT+8), Sam Altman told OpenAI staff that the company has no authority to decide how the US Defense Department uses its AI software in operations. According to a person familiar with the meeting, the Defense Department said it would listen to OpenAI’s technical expertise on applications, but it did not want the company to judge whether specific military actions were good or bad ideas. Altman also suggested this boundary may have contributed to tension between the Pentagon and Anthropic.

The Tuesday meeting was Altman’s first employee Q&A after OpenAI reached a late-Friday agreement allowing Pentagon deployment of OpenAI models on a classified network, a gap of about 4 days. The deal followed a Pentagon-Anthropic standoff in which Anthropic sought limits against mass surveillance of Americans and against deployment of fully autonomous weapons. The report also says Anthropic raised questions about technology use in the raid to capture Venezuelan President Nicolas Maduro, though Anthropic denied discussing specific operations with the Defense Department.

Altman had said the Pentagon arrangement matched OpenAI principles barring domestic mass surveillance and requiring human responsibility in the use of force, including autonomous weapon systems. He later called the rushed deal opportunistic and sloppy and said OpenAI was adding terms to clarify principles, including preventing domestic surveillance uses and preventing intelligence agencies such as the National Security Agency from relying on OpenAI services. Altman also said he is still pressing the Pentagon to remove Anthropic’s supply-chain-risk designation, described as not previously applied to a US company.