Trump may be forced to reveal secret rules feds use for AI safety testing 特朗普或被迫公开联邦政府用于AI安全测试的秘密规则

TL;DR

• Protect Democracy filed a lawsuit against four federal agencies demanding transparency around the Trump administration's secret voluntary safety review framework for frontier AI models
• The framework, completed August 3, allows government review of prerelease AI models with reduced safeguards, but no details about its criteria, participants, or legal authority have been made public
• OpenAI has reportedly negotiated a private agreement limiting distribution of its cutting-edge models to government-vetted partners only
• Critics allege the process could enable corruption, with officials potentially favoring AI firms aligned with Trump while retaliating against companies like Anthropic
• California's proposed SB 813 offers a transparent alternative with independent organizations setting public AI safety standards

Why It Matters

This lawsuit strikes at the heart of AI governance accountability, as the executive branch now controls which frontier models reach the public without congressional oversight or public scrutiny. For AI practitioners and researchers, the lack of transparency around safety review criteria creates uncertainty about compliance requirements and could shape market access for next-generation models. The tension between national security concerns and democratic accountability will influence how AI regulation evolves in the United States and globally.

Technical Details

• The framework operates through the Center for AI Standards and Innovation, which reviews prerelease models with "reduced or removed safeguards" to evaluate national security-related capabilities and risks
• A parallel clearinghouse called GOLD EAGLE was launched in July to flag cybersecurity vulnerabilities across sectors, relying on unidentified industry partners
• The executive order authorizes a "classified benchmarking process" for assessing advanced AI capabilities, adding another layer of opacity to the review mechanism
• The lawsuit demands disclosure of unclassified procedural and contractual architecture, including the framework's text, participation terms, participant identities, and criteria for granting or withholding model access
• California's SB 813 proposes independent organizations establishing transparent AI safety baselines, benchmarks, and evaluation methodologies as a contrast to the federal approach

Industry Insight

AI companies should prepare for increasing regulatory scrutiny and potential mandatory disclosure requirements as transparency lawsuits gain traction; proactive engagement with clear safety protocols will be advantageous. The lack of published review criteria creates compliance risk for frontier model developers, who may face unpredictable government demands regarding model distribution and safety testing. The contrast between California's transparent legislative approach and the federal secret framework suggests a multi-track regulatory landscape where companies may need to navigate differing standards across jurisdictions.

TL;DR

• 特朗普政府秘密建立前沿AI模型安全审查框架，四个联邦机构被起诉要求公开
• 非营利组织Protect Democracy指控审查流程缺乏透明度，未向公众或国会披露细节
• OpenAI已与联邦政府达成私人协议，限制其尖端AI模型仅向政府审核的合作伙伴分发
• 加州SB 813法案提出建立独立、透明的AI安全标准流程，与政府做法形成对比
• 诉讼要求9月30日前公开框架文本、参与方身份及准入标准等关键信息

为什么值得看

本文揭示了美国政府在AI治理领域的透明度危机，涉及行政权力扩张与公众知情权的冲突。对AI从业者而言，这直接关系到模型发布合规要求及与政府合作的风险评估，也反映了全球AI监管走向透明化与独立化的趋势。

技术解析

• 审查框架架构：特朗普政府于8月3日宣布完成自愿性AI模型预发布审查框架，由四个联邦机构（国家网络总监办公室、科学技术政策办公室、财政部、商务部）共同参与，但框架文本、参与标准及法律授权均未公开。
• GOLD EAGLE信息交换中心：7月由白宫推出的行业合作平台，依赖企业伙伴帮助各机构识别网络安全漏洞，但未披露参与公司名称、条款及法律依据。
• OpenAI私人协议：OpenAI已与联邦政府达成私下协议，限制其前沿AI模型仅向政府审核的合作伙伴分发，引发对监管俘获和公平竞争的担忧。
• 加州SB 813法案：加州正在审议的立法提案，主张由独立组织制定AI安全基准，确保评估标准、方法论和基准测试对公众透明，与政府秘密框架形成鲜明对比。
• Anthropic Mythos 5事件：今年夏天政府以安全风险为由阻止Anthropic发布Mythos 5模型，成为推动建立审查框架的直接导火索。

行业启示

• 监管透明度将成为AI行业核心议题：政府秘密审查框架可能影响模型发布节奏和市场准入，企业需关注法律进展并评估合规风险，同时推动监管流程公开化。
• 独立第三方安全评估机构将崛起：加州SB 813代表的透明化路径暗示，未来AI安全标准可能由独立组织制定，企业应提前布局第三方审计和认证能力。
• 政企关系面临重新定义：OpenAI等公司与政府的私人协议引发"监管俘获"担忧，行业需建立更公平的参与机制，避免少数巨头垄断政策话语权。