内容来源：https://www.engadget.com/ai/openai-strikes-a-deal-with-the-defense-department-to-deploy-its-ai-models-054441785.html?src=rss

内容总结：

美国国防部与人工智能公司OpenAI于近日达成合作协议，OpenAI将为其内部网络部署人工智能模型。该公司首席执行官萨姆·阿尔特曼在社交媒体上透露，协议中包含禁止将技术用于国内大规模监控及确保人类对使用武力（包括自主武器系统）负责两项核心安全原则。

此前，美国政府因另一家人工智能公司Anthropic拒绝移除其AI模型中的安全限制措施而终止合作，并威胁将其列为"供应链风险"。Anthropic公司坚持立场，声明不会在禁止国内大规模监控和全自主武器两大原则问题上妥协。

值得关注的是，OpenAI与国防部的协议中虽设有安全机制，但其技术部署范围将限于云端网络。目前OpenAI正与亚马逊云服务（AWS）建立合作关系，这可能为其未来接入政府采用的亚马逊云平台创造条件。

分析人士指出，此次合作反映出美国军方在推进人工智能军事应用过程中，正试图在技术能力与伦理约束之间寻求平衡。

中文翻译：

OpenAI与美国国防部达成协议，将部署其人工智能模型

OpenAI首席执行官萨姆·奥尔特曼在X平台透露，公司已与国防部达成协议，将在该机构的内部网络中部署其人工智能模型。他在发文中强调，OpenAI最重要的两项安全原则是"禁止国内大规模监控"以及"确保人类对使用武力的责任，包括自主武器系统"。奥尔特曼称这些原则已被写入与国防部的协议，并特别使用政府偏好的"战争部"称谓指代该机构，表示对方已承诺遵守这些条款。

该协议达成之际，正值唐纳德·特朗普总统下令所有政府机构停止使用Claude及Anthropic其他服务之后。此前美国国防部长皮特·赫格塞斯曾威胁称，若Anthropic继续拒绝移除其人工智能的安全护栏——这些限制旨在防止该技术被用于针对美国公民的大规模监控及全自主武器——将把该公司列为"供应链风险"。

目前尚不清楚政府为何选择与同样设有安全护栏的OpenAI合作。奥尔特曼表示，公司正要求政府为所有合作的人工智能企业提供同等条款。主管外援、人道事务及宗教自由的副国务卿杰里米·卢因在X平台指出，战争部的合同中"援引了现有法律授权，并包含双方认可的安全机制"。此前已签署协议在战争部机密系统部署Grok模型的xAI公司，也与OpenAI共同接受了这些条款。卢因称这与"当初向Anthropic提出但遭拒绝的折中方案完全相同"。

自2024年开始与美国政府合作的Anthropic公司始终拒绝向赫格塞斯妥协。就在奥尔特曼宣布OpenAI协议前数小时，该公司再次发表声明重申立场："战争部的任何恐吓或惩罚都无法改变我们在国内大规模监控和全自主武器问题上的立场，我们将在法庭上挑战任何供应链风险认定。"

奥尔特曼在X平台补充说明，OpenAI将建立技术保障措施以确保模型行为符合规范，并称这也符合战争部的期望。公司将派遣工程师与国防部协作"确保模型安全性"，且仅通过云网络进行部署。《纽约时报》指出，OpenAI尚未接入政府使用的亚马逊云服务，但随着该公司近日宣布与亚马逊建立合作伙伴关系，允许企业客户通过亚马逊云服务运行其模型，这一现状可能即将改变。

（附奥尔特曼推文译文：今晚，我们与战争部达成协议，将在其机密网络中部署我们的模型。在所有接触中，战争部始终展现出对安全的高度重视，以及通过合作实现最佳成果的意愿。人工智能安全与广泛部署……——萨姆·奥尔特曼 2026年2月28日）

英文来源：

OpenAI strikes a deal with the Defense Department to deploy its AI models
The agreement comes after the US government soured on Anthropic for refusing to remove its AI's safety guardrails.
OpenAI has reached an agreement with the Defense Department to deploy its models in the agency’s network, company chief Sam Altman has revealed on X. In his post, he said two of OpenAI’s most important safety principles are “prohibitions on domestic mass surveillance and human responsibility for the use of force, including for autonomous weapon systems.” Altman claimed the company put those principles in its agreement with the agency, which he called by the government’s preferred name of Department of War (DoW), and that it had agreed to honor them.
The agency has closed the deal with OpenAI, shortly after President Donald Trump ordered all government agencies to stop using Claude and any other Anthropic services. If you’ll recall, US Defense Secretary Pete Hegseth previously threatened to label Anthropic “supply chain risk” if it continues refusing to remove the guardrails on its AI, which are preventing the technology to be used for mass surveillance against Americans and in fully autonomous weapons.
It’s unclear why the government agreed to team up with OpenAI if its models also have the same guardrails, but Altman said it’s asking the government to offer the same terms to all the AI companies it works with. Jeremy Lewin, the Senior Official Under Secretary for Foreign Assistance, Humanitarian Affairs, and Religious Freedom, said on X that DoW “references certain existing legal authorities and includes certain mutually agreed upon safety mechanisms” in its contracts. Both OpenAI and xAI, which had also previously signed a deal to deploy Grok in the DoW’s classified systems, agreed to those terms. He said it was the same “compromise that Anthropic was offered, and rejected.”
Anthropic, which started working with the US government in 2024, refused to bow down to Hegseth. In its latest statement, published just hours before Altman announced OpenAI’s agreement, it repeated its stance. “No amount of intimidation or punishment from the Department of War will change our position on mass domestic surveillance or fully autonomous weapons,” Anthropic wrote. “We will challenge any supply chain risk designation in court.”
Altman added in his post on X that OpenAI will build technical safeguards to ensure the company’s models behave as they should, claiming that’s also what the DoW wanted. It’s sending engineers to work with the agency to “ensure [its models’] safety,” and it will only deploy on cloud networks. As The New York Times notes, OpenAI is not yet on Amazon cloud, which the government uses. But that could change soon, as company has also just announced forming a partnership with Amazon to run its models on Amazon Web Services (AWS) for enterprise customers.
Tonight, we reached an agreement with the Department of War to deploy our models in their classified network.
In all of our interactions, the DoW displayed a deep respect for safety and a desire to partner to achieve the best possible outcome.
AI safety and wide distribution of…— Sam Altman (@sama) February 28, 2026