美国 AI 政策

涉华风投、私募股权交易中所有AI相关标的均需禁止投资，不再仅限前沿模型

联邦民用机构及其供应商不得在美国政府IT系统中部署或保留DeepSeek产品或基于其开发的第三方服务。

在美国国土安全部体系内承接业务或部署系统的公司不得使用 DeepSeek 模型。

使用DeepSeek模型或服务的联邦民用机构须立即停止相关使用并寻找替代方案。

被列名实体将被冻结美国境内资产，且无法与美国人士进行任何交易，融资和合作渠道即刻中断

联邦民用机构不得采购或使用DeepSeek的AI产品和服务，相关合同和工具需立即停用

向美国联邦机构销售或交付服务的供应商需排查并移除其产品中集成的 DeepSeek 技术或衍生组件。- 基于 DeepSeek 微调、蒸馏或封装的第三方服务同样被禁，涉及美国政府业务的企业需评估模型替换方案。- 中国 AI 模型在美联邦市场面临进一步准入限制，依赖 DeepSeek 的商业化路径需重新评估。

向美国联邦政府或其供应商提供产品或服务的企业，须确认其技术栈中不包含 DeepSeek 或其衍生模型。

联邦文职机构供应链中的 DeepSeek 及其衍生产品面临移除要求，相关集成商和供应商需排查并替换依赖该模型的产品。- 基于开源中国模型构建的企业产品在进入美国政府市场时受到采购限制，需重新评估目标客户与模型选型。- 中美在 AI 出口与采购上的双向管制收紧，为在美运营的中国 AI 企业及在 ASEAN 布局的厂商带来合规审查压力。

禁止向美国联邦民事行政部门供应 DeepSeek 原厂或基于其架构的第三方软硬件与服务，相关采购合同即时受阻。

向美国联邦政府供货的企业需确认供应链中不含 DeepSeek 组件，否则订单资格受影响

向美国政府或联邦承包体系提供产品的公司需排查并移除基于DeepSeek实现的组件，否则面临合同合规风险。- 基于DeepSeek模型做二次开发或微调的第三方服务商被明确纳入禁令范围，衍生实现同样不可用于联邦业务。- 涉及中美AI供应链的企业需评估美国客户市场的准入限制，并调整技术栈以符合采购要求。

向联邦民用机构供货的软件供应商须排查产品内含的DeepSeek等开源权重模型，并证明供应链合规，否则面临订单风险

面向美国联邦政府提交的文件、投标书和法庭文书需将“AI”表述改为“超级智能/SI”，以符合行政令要求

美国AI公司需调整模型发布流程，确保美国机构优先审查

AI公司需调整模型发布流程，确保美国政府获得优先测试权

Obligations: Data center operators must increase community participation in project approvals and cannot use confidentiality agreements. They must also comply with stricter environmental and energy efficiency standards. • Opportunities: Data centers investing in clean energy may receive priority for grid access. • Deadlines: The framework includes immediate administrative measures and proposed legislation for the 2027 state legislature.

Obligations: AI companies may face legal scrutiny and potential penalties if found to have coordinated to limit AI development speed, violating antitrust laws. • Opportunities: The case highlights the importance of transparent communication and compliance with antitrust regulations in the AI sector. • Deadlines: The court case is ongoing, with the possibility of a collective class-action suit being certified, which could impact the companies' operations and public perception.

必须为AI模型配备‘kill switch’功能并设立独立审计员； - 需在两个月内配合专家小组提供相关建议； - 需遵守加州AI安全监管要求，否则面临合规风险。

Obligations: AI companies may face increased regulatory scrutiny and potential legal action over data usage and its impact on news publishers. • Opportunities: Companies may need to develop more transparent data practices and collaborate with news organizations to mitigate negative impacts. • Deadlines: The FTC consultation period is open for public comments until a specified date, after which the agency will consider feedback before potentially adopting new guidelines.

Obligations: AI companies operating in California may face new regulatory requirements, including potential mandatory security audits by independent third parties. • Opportunities: Companies may benefit from the development of a clearer regulatory framework for AI safety, potentially reducing uncertainty. • Deadlines: Companies must prepare for the upcoming two-month timeline for the expert panel to submit its recommendations and the subsequent legislative process.

Obligations: US AI companies may face increased scrutiny and potential regulatory constraints on their development speed.

Obligations: Companies planning data center projects in San Jose must undergo a more rigorous public review process and provide detailed disclosures on energy consumption, water usage, and environmental impact assessments. • Opportunities: Successful companies that can demonstrate responsible development and community benefits may gain public support and approval for their projects. • Deadlines: The city is currently consulting the public, and while no specific deadline is mentioned, companies must align with the city's proposed review framework before project approval.

OpenAI 将调整研发策略，优先考虑安全评估，可能放缓前沿技术迭代速度。

义务：前沿AI开发者需遵守联邦框架下的统一安全要求，并接受独立审计。• 机遇：通过建立统一的安全标准和协作框架，可提升AI发展的安全性，增强行业信心。• 时间节点：呼吁立即行动，但具体法规和框架尚未出台。

Obligations: AI companies, particularly those in the US, are under pressure to demonstrate responsible development and transparency regarding their AI models.

企业需掌握自身独有的隐性知识，构建持续学习闭环，减少对单一模型供应商的依赖。

企业需确保其AI研究符合人类控制与普惠原则，避免追求无意义或有害的超智能。• 企业应保留对自身独特知识的控制权，并构建自主的持续学习循环，不依赖单一模型提供商。• 微软将发布“行为准则”，为行业提供明确的指导方向。

Obligations: AI data center operators may face new environmental health regulations if the plan is adopted, requiring compliance with stricter air quality standards. • Opportunities: Companies investing in AI data centers in the US could benefit from a streamlined regulatory process if the plan is implemented. • Deadlines: The plan's adoption depends on the president's decision, with no specific timeline provided in the report.

Obligations: Companies receiving grants must disclose research findings on AI risks to the public and regulatory bodies. • Opportunities: Funding provides resources for AI safety research, potentially leading to safer AI development. • Deadlines: Recipients must submit quarterly progress reports and final research outcomes within 18 months of grant approval.