Anthropic 發布強大的網絡安全 AI 模型
Anthropic Unveils Powerful AI Model for Cybersecurity
2026年4月，人工智慧研究公司Anthropic推出了Claude Mythos Preview，這是一款功能強大、具有無與倫比編碼和推理能力的新型AI模型。
In April 2026, AI research firm Anthropic introduced Claude Mythos Preview, a powerful new AI model with unmatched coding and reasoning abilities.
與以往的模型不同，Mythos具有識別並利用軟體漏洞的危險能力。
Unlike previous models, Mythos possesses the dangerous capability to identify and exploit software vulnerabilities.
由於這些風險，Anthropic並未公開發布該模型。
Due to these risks, Anthropic has not released it publicly.
相反，他們啟動了「玻璃翼計畫」(Project Glasswing)，這是一項封閉式計畫，僅授予Google和Microsoft等特定科技巨頭對該模型的受控存取權。
Instead, they launched Project Glasswing, a gated initiative granting select tech giants, like Google and Microsoft, controlled access to the model.
其目標是利用Mythos的防禦優勢，在攻擊者發現漏洞之前修復關鍵錯誤。
The goal is to use Mythos's defensive strengths to fix critical bugs before attackers find them.
在內部測試中，Mythos發現了數千個高嚴重性缺陷，包括幾十年來一直未被察覺的漏洞。
During internal testing, Mythos uncovered thousands of high-severity flaws, including decades-old vulnerabilities that had escaped detection until now.
雖然一些批評者對其行銷策略提出質疑，但其他人則認為這是對AI驅動的網路威脅時代的關鍵回應。
While some critics question the marketing strategy, others see it as a vital response to the era of AI-driven cyber-threats.
Anthropic相信，透過賦予防禦者代理型AI——即能自主執行複雜任務的系統——我們能夠達到更具韌性的安全平衡。
Anthropic believes that by empowering defenders with agentic AI—systems that autonomously execute complex tasks—we can reach a more resilient security equilibrium.
隨著聯邦官員討論對金融穩定性的風險，全世界都在關注Mythos是否真的會將平衡傾向於保護數位基礎設施的一方。
As federal officials discuss the risks to financial stability, the world watches to see if Mythos will indeed shift the balance in favor of those protecting the digital infrastructure.
