Expired50% confidenceOpinionExact time
Sonnet 5 was deliberately undertrained on cybersecurity tasks in order to maintain an ASL-2 rating and bypass regulatory scrutiny for a faster release
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/20/2026
First Seen
Valid until: 10/18/2026(expired)
Sources
Related Claims
UnverifiedAnthropic has enabled cyber safeguards for Sonnet 5 by default68% similarUnverifiedSafety classifiers are highly sensitive to request phrasing; asking the model directly to find vulnerabilities is easily intercepted and routed to Opus 4.8, whereas neutral descriptions tend not to trigger restrictions.55% similarUnverifiedH5ST (H5 Security Token) is a front-end security signing mechanism independently developed by JD.com and is a core component of the web-side anti-crawler system.53% similarUnverifiedHermes设计了五层安全保护机制，包括测试套件全通过、提示词大小限制、缓存机制保护、核心意图锁定和人类审查关卡53% similarUnverified2024年安全研究者发现的ASCII走私技术利用Unicode中人眼不可见但AI可解读的特殊字符隐藏恶意指令，该技术已在Microsoft Copilot等产品中被验证有效53% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/571458API
curl https://kongchang.com/api/v1/knowledge/claims/571458MCP
get_claim(id=571458)