科技巨頭推出 AI 工具以挖掘軟體安全漏洞
Tech Giants Launch AI Tool to Find Software Security Flaws
2026年4月7日，一項名為「玻璃之翼計畫」(玻璃之翼計畫ㄅㄛㄌㄧㄓㄨㄧㄥㄐㄧㄏㄨㄚˋ)的開創性倡議正式啟動，旨在加強全球軟體安全。
On April 7, 2026, a groundbreaking initiative called Project Glasswing was launched to fortify global software security.
該計畫由Anthropic領軍，與Google、Microsoft和Amazon等科技巨頭合作，推出了Claude Mythos Preview，這是一款功能強大的AI模型，旨在檢測隱藏的安全漏洞。
Led by Anthropic in partnership with tech giants like Google, Microsoft, and Amazon, the project introduces Claude Mythos Preview, a powerful AI model engineered to detect hidden security flaws.
與傳統工具不同，Mythos擅長掃描龐大的代碼庫，並已發現了隱藏數十年的弱點，例如OpenBSD中一個存在了27年的臭蟲。
Unlike traditional tools, Mythos excels at scanning massive codebases, having already uncovered vulnerabilities that remained hidden for decades, such as a 27-year-old bug in OpenBSD.
鑑於該工具存在被濫用的潛在風險，Anthropic限制了對其的存取權限，僅提供給基礎設施關鍵夥伴。
Recognizing the potential for misuse, Anthropic is restricting access to the tool, providing it only to essential infrastructure partners.
透過投入1億美元的支持，該倡議凸顯了整個產業向防禦性AI轉型的趨勢。
By committing $100 million in support, this initiative highlights an industry-wide pivot toward defensive AI.
隨著網路安全威脅不斷演變，玻璃之翼計畫表明，以人為主的安全性已不足以管理我們日益複雜的數位生態系統。
As cybersecurity threats evolve, Project Glasswing demonstrates that human-led security is no longer enough to manage our increasingly complex digital ecosystem.
