來源The Next Web (TNW)•較早收集於 86m
美國政府下令 Anthropic 召回 Fable 5 與 Mythos 5 模型
#ai-regulation#national-security#model-safetyfable-5-and-mythos-5anthropicfable 5mythos 5
💡美國政府首次強制召回商業 AI 模型，為產業設立了重大的監管先例。
⚡ 30 秒速覽
有什麼變化
Anthropic 於 6 月 12 日收到政府指令，要求暫停 Fable 5 與 Mythos 5。
為什麼重要
此舉為 AI 部署建立了重大的監管先例，顯示政府願意直接干預商業模型的可用性。這為依賴專有模型進行長期基礎設施建設的開發者帶來了極大的不確定性。
下一步行動
檢視您對特定專有模型的依賴程度，並實施多模型備援策略，以減輕監管突然下架所帶來的風險。
誰應關注：Founders & Product Leaders
關鍵要點
- •Anthropic 於 6 月 12 日收到政府指令，要求暫停 Fable 5 與 Mythos 5。
- •這是美國政府首次強制要求商業 AI 模型下線的已知案例。
- •此次召回明確是基於國家安全權限所發布。
🧠 深度解析
背景與延伸：來自公開資料，非原文內容。引用 12 個來源。
🔑 增強重點摘要
- •The US government's directive, issued by the Commerce Department, specifically mandated Anthropic to suspend access to Fable 5 and Mythos 5 for all foreign nationals, including its own foreign-national employees, both inside and outside the US.
- •Anthropic was compelled to disable both models for all customers globally because it lacked the real-time capability to differentiate between foreign and domestic users, ensuring compliance with the export control directive.
- •The government's national security concern reportedly stems from a 'jailbreak' method discovered for Fable 5, which Anthropic contends only exposed minor, previously known vulnerabilities that are also present in other publicly available AI models.
- •Fable 5, launched just three days before the recall, was the public-facing version of Anthropic's advanced 'Mythos-class' model, while Mythos 5, with fewer cyber safeguards, was intended for vetted cybersecurity and critical infrastructure operators.
- •This recall is part of ongoing tensions between Anthropic and Washington, which previously included the Pentagon blacklisting Anthropic as a national security supply chain threat, even as the NSA continued to utilize Claude on classified networks.
🛠️ 技術深入
- Fable 5 is designed with safety classifiers that redirect flagged requests, particularly those related to cybersecurity misuse, to the less capable Claude Opus 4.8 model.
- Mythos 5 utilizes the same core underlying model as Fable 5 but operates with lifted cyber safeguards, making it accessible only to a restricted group of vetted cybersecurity defenders and critical infrastructure operators.
- Anthropic implemented "strong safeguards" in Fable 5, including safety classifiers, to detect and prevent misuse for cybersecurity-related tasks and jailbreak attempts.
- Mythos-class models are noted for their proficiency in identifying and exploiting software vulnerabilities, capable of generating working exploits from newly disclosed vulnerabilities within hours or even minutes.
- Prior to Fable 5's launch, Anthropic conducted thousands of hours of red-teaming on its safeguards, involving the US government, the UK AISI, and third-party organizations, concluding that its safeguards were significantly more effective than those of previous models and that no universal jailbreak had been found.
- The specific jailbreak technique cited by the government reportedly identified minor, known vulnerabilities that Anthropic states can also be discovered by other publicly available AI models without requiring a special bypass.
🔮 前景展望基於引用來源的 AI 分析
The recall sets a significant precedent for government intervention in commercial AI model deployment.
This marks the first known instance of the US government compelling a commercial AI product to be taken offline, signaling a new era of direct federal oversight on frontier AI technologies.
AI developers may face increased pressure for pre-release government vetting and nationality-based access controls.
The directive's focus on restricting access for foreign nationals and Anthropic's subsequent global shutdown suggest future regulatory requirements for identity verification and early government access to advanced AI models.
The incident will likely intensify the debate around AI governance, transparency, and the balance between national security and innovation.
Anthropic publicly expressed disagreement with the government's decision, advocating for a transparent, fair, and technically grounded statutory process for blocking unsafe deployments, which highlights the ongoing tension in AI regulation.
⏳ 時間線
2021-01
Anthropic is founded by former OpenAI researchers, with a focus on AI safety.
2023-03
Claude 1, Anthropic's first public AI model, is launched.
2024-03
Anthropic launches the Claude 3 model family (Haiku, Sonnet, and Opus).
2025-05
The Claude 4 generation, including Opus 4 and Sonnet 4, is released.
2026-06-02
President Trump signs an executive order on AI and national security, calling for benchmarking advanced cyber capabilities of AI models and requesting voluntary early access to frontier models from developers.
2026-06-09
Anthropic launches Fable 5 as its first public 'Mythos-class' model, with Mythos 5 restricted to vetted users.
2026-06-12
The US government issues an export control directive to Anthropic, ordering the suspension of Fable 5 and Mythos 5 access for foreign nationals, leading to a global shutdown.
📎 來源 (12)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
📰
AI 週報
閱讀本週精選 AI 大事摘要 →
👉相關動態
AI 策展新聞聚合。所有內容版權歸原始發布者所有。
原始來源： The Next Web (TNW) ↗
每週電子報
每週一封，可隨時退訂。