Trump Administration Lifts Restrictions on Anthropic’s Claude Models After Cybersecurity Alarm 特朗普政府在网络安全警报后解除对Anthropic Claude模型的限制

TL;DR

• The Trump administration has lifted restrictions on Anthropic's Claude Fable 5 and Mythos 5 models, ending a ban triggered by cybersecurity concerns regarding vulnerability exploitation.
• Claude Fable 5 is now widely available, while the more powerful Mythos 5 model is restricted to a select group of U.S.-based organizations approved by the federal government.
• The initial ban was prompted by Amazon security researchers discovering methods to bypass Fable 5's safeguards to find software vulnerabilities, raising national security alarms.
• Competitor OpenAI is simultaneously restricting its GPT-5.6 Sol model to government-approved customers under similar executive orders establishing a 30-day vetting framework for advanced AI systems.

Why It Matters

This development marks a significant shift in how the U.S. government regulates frontier AI capabilities, moving from broad restrictions to a targeted, risk-based approval process for specific high-power models. For AI practitioners and companies, it highlights the increasing importance of national security compliance and the necessity of proactive engagement with federal oversight bodies like the Commerce Department. It also signals that even "voluntary" frameworks may effectively become mandatory for deploying state-of-the-art models, impacting global access and competitive dynamics.

Technical Details

• Model Specifications: Anthropic released two distinct tiers: Claude Fable 5, which has unrestricted public access, and Mythos 5, a more powerful model limited to vetted U.S. entities. OpenAI introduced GPT-5.6 Sol, similarly restricted to government-approved customers.
• Security Concerns: The regulatory intervention was driven by reports that Amazon security researchers identified methods to bypass safeguards in Claude Fable 5, allowing it to discover and potentially exploit software vulnerabilities.
• Regulatory Framework: A recent executive order established a mechanism for the federal government to vet national security risks of advanced AI systems for up to 30 days prior to public release, although the framework's full development is still ongoing.
• Implementation: The restrictions initially blocked foreign nationals from using the models, forcing Anthropic to take products down immediately after unveiling them, demonstrating the rapid enforcement capability of the new oversight regime.

Industry Insight

AI developers must integrate robust security auditing and red-teaming into their pre-release pipelines, particularly focusing on preventing vulnerability discovery and exploitation capabilities, to avoid regulatory shutdowns. Companies should anticipate stricter, de facto mandatory compliance requirements for frontier models, necessitating early dialogue with government agencies to secure approval for broader distribution. The divergence in access levels between models suggests a future where AI capabilities are stratified by security clearance, creating new barriers to entry and potential geopolitical fragmentation in AI adoption.

TL;DR

• 特朗普政府解除对Anthropic最新Claude模型（Fable 5和Mythos 5）的禁令，结束了因网络安全担忧导致的数周限制。
• 此前美国商务部禁止外国人使用这两款模型，原因是亚马逊安全研究人员发现Fable 5存在绕过安全防护以利用软件漏洞的方法。
• Anthropic的Mythos 5模型虽恢复可用，但仅限美国政府批准的少数美国组织访问；Fable 5则向公众广泛开放。
• 竞争对手OpenAI也响应行政命令，将其新模型GPT-5.6 Sol的限制访问权限仅限于政府批准的客户。
• 这一调整基于特朗普签署的AI监管行政令，该令状允许联邦政府在发布前对先进AI系统进行长达30天的国家安全风险评估。

为什么值得看

本文揭示了美国政府对顶尖AI模型实施实质性国家安全审查的最新进展，标志着AI监管从自愿框架向强制性安全审查过渡的关键节点。对于AI从业者和企业而言，理解这种“政府审批+分级访问”的新合规模式，是确保产品在全球市场（尤其是美国市场）顺利发布的必要条件。

技术解析

• 模型版本与访问策略：Anthropic发布了Claude Fable 5（广泛可用）和Mythos 5（受限访问）。Mythos 5被描述为最强大的模型，专门针对软件漏洞发现能力进行了优化，但也因此面临更高的安全风险。
• 安全漏洞根源：限制措施源于亚马逊云科技（Anthropic的主要云提供商）安全研究人员的报告，指出Fable 5存在被绕过安全防护的方法，可能被恶意黑客用于发现和利用软件漏洞。
• 监管框架执行：依据特朗普总统签署的行政令，联邦政府有权在模型公开发布前进行最高30天的国家安全审查。尽管参与初期被描述为自愿，但实际执行中已表现出强制性的行业影响，迫使Anthropic和OpenAI同步调整发布策略。
• 竞品动态：OpenAI同步推出GPT-5.6 Sol，并同样采取仅向政府批准客户开放的策略，表明主要AI厂商均在遵循同一套国家安全合规标准。

行业启示

• 安全合规成为核心竞争力：AI模型的发布不再仅取决于技术性能，更取决于其通过国家安全审查的能力。开发者需将“可验证的安全性”和“漏洞防御机制”置于与模型能力同等重要的地位。
• 政企协作新模式：政府通过行政令深度介入AI产品的生命周期管理，形成了“政府审批-厂商受限发布-用户分级访问”的新型生态。企业需建立专门的政府关系与安全合规团队以应对此类监管要求。
• 全球市场分化风险：由于限制主要针对外国用户或特定群体，这可能加剧AI技术的国界化趋势。跨国AI企业需针对不同司法管辖区制定差异化的部署和访问控制策略，以平衡创新速度与合规成本。