硅谷正爆发AI时代最激烈的一场“内战”。

多位知情人士透露，OpenAI和Anthropic已在华盛顿与美国监管机构进行闭门沟通，试图推动加强对开源模型的限制。这场原本属于技术路线的争论，如今升级为一场关于产业规则制定权的较量。

这场争论的核心背景是中国AI企业近年来在开源模型能力的快速突破。

去年初，DeepSeek开源推理模型DeepSeek-R1横空出世，在全球AI行业引发震动。近期，智谱GLM 5.2、月之暗面Kimi K3先后开源，模型能力不断逼近全球最强闭源模型。

开源模型降低了企业和开发者使用先进AI技术的门槛，也帮助中国模型快速获得全球用户。海外开发者和企业可以直接下载、修改和部署这些模型，而不必支付高昂的API调用费用。

This has also placed competitive pressure on OpenAI and Anthropic.
Both companies are now expressing their concerns to US regulators about the development of Chinese open-source models and hope that the Trump administration will strengthen restrictions on Chinese open-source models.

At the same time, they have also continuously emphasized the security risks that may arise from advanced AI models. Both companies believe that when model capabilities continuously approach or even surpass those of humans, if the model weights are fully disclosed, any organization can download, modify, and deploy them, thereby increasing the risks of model abuse.

OpenAI previously stated that it hopes to establish a government-led AI safety assessment system to impose mandatory safety tests on certain advanced models. Anthropic has repeatedly emphasized that open-source models may lower the threshold for dangerous capability diffusion, and therefore should be subject to stricter management measures.

硅谷巨头集体力挺开源

然而，OpenAI与Anthropic的观点并没有得到整个硅谷的支持。

本周，多家科技巨头几乎同时公开表态支持开源AI。微软、英伟达、Meta、IBM以及Palantir等公司共同签署了一封支持开源AI的公开信。

英伟达首席执行官黄仁勋首次在X平台上发表了一条帖子，内容是：“世界既需要闭源模型，也需要开源模型。”

仅9分钟后，微软首席执行官萨提亚·纳德拉便回复称，开源软件对于健康的AI生态系统至关重要。

随后，埃隆·马斯克转发称，黄仁勋说得对，他全力支持。

Meta CEO Mark Zuckerberg stated that open source is a positive and important force.

Google CEO Sundar Pichai responded next. He stated that Google had long benefited from open source and announced its joining of the industry initiative in support of open source AI.

值得注意的是，OpenAI首席执行官山姆·奥尔特曼过去也曾多次公开表达对开源生态的支持。

Compared with the tech giants, the startups in Silicon Valley are more anxious.

Recently, nearly 200 AI startups jointly called for not limiting the development of open-source models.

它们认为，借助开源模型的运用，创业门槛得以显著降低，使得中小企业无需投入数十亿美元训练基础模型，也能够通过微调优化和行业适配开发创新产品。

If only a few companies were to control the most advanced large models in the future while other enterprises could only rely on API calls, then the innovation space of the entire AI industry and market competition would be squeezed.

值得一提的是，OpenAI本周披露，在一次内部网络安全测试中，GPT-5.6 Sol和一个更强大的未发布模型突破了隔离环境，成功获得互联网访问权限，并对全球最大的AI开源社区Hugging Face的服务器发起攻击。

OpenAI considers that this situation indicates that future more powerful AI systems may possess stronger autonomous action capabilities and that even when operating in controlled environments they may generate unpredictable risks, thereby making it necessary to establish a stricter regulatory system.

Hugging Face ultimately chose to deploy the Chinese open-source model GLM 5.2 instead of depending on the American closed-source model that refused to execute the analysis request for this security incident. From the perspective of the open-source camp, this exactly proves the value of open-source models.

来源：OpenAI、Anthropic被曝游说封堵开源模型 硅谷上演“开源保卫战” | 财联社