Anthropic 發布了其 AI 模型更安全、更便於公眾使用的版本
Anthropic releases a safer, public-friendly version of its AI model
Updated at: June 11, 2026 at 04:20 AM
在2026年6月9日，AI公司Anthropic推出了Claude Fable 5，這是其強大的Mythos模型的新版本，專為大眾設計。
On June 9, 2026, AI company Anthropic introduced Claude Fable 5, a new version of its powerful Mythos model designed for the general public.
這次發布代表了在平衡高效能人工智慧與必要安全性方面邁出的一大步。
This launch represents a major step in balancing high-performance artificial intelligence with essential safety.
此前，原始的Mythos模型因具備識別危險網路安全漏洞的能力，僅限於特定的研究小組使用。
Previously, the original Mythos model was restricted to specific research groups because of its ability to identify dangerous cybersecurity vulnerabilities.
為了填補這一空白，Anthropic開發了Fable 5，它使用與Mythos相同的先進架構，但包含了嚴格的安全護欄。
To bridge this gap, Anthropic created Fable 5, which uses the same advanced architecture as Mythos but includes strict safety guardrails.
當AI檢測到高風險請求（例如涉及毒素、生物學或網路威脅的請求）時，它會自動將任務轉移至較不複雜的Claude Opus 4.8模型。
When the AI detects a high-risk request—such as those involving toxins, biology, or cyber threats—it automatically shifts the task to the less complex Claude Opus 4.8 model.
Anthropic指出，此系統對不到5%的用戶產生干擾，確保大多數人能在無問題的情況下享受完整效能。
Anthropic notes that this system interrupts fewer than 5% of users, ensuring most people enjoy full performance without issues.
這種雙軌策略——為大眾提供過濾版本，為授權合作夥伴提供無限制版本——凸顯了科技界在提供尖端運算能力同時維持安全性方面持續的掙扎。
This dual-track strategy—offering a filtered version for the public and an unrestricted one for authorized partners—highlights the ongoing struggle in the tech world to provide cutting-edge power while maintaining safety.
這些介面層級的保護措施是否足以遏制先進AI，仍是專家間激烈辯論的主題。
Whether these interface-level protections are enough to contain advanced AI remains a topic of intense debate among experts.
