金色财经|7月 03, 2026 01:21
**[Anthropic Releases Security Measures for Fable 5 Model and AI "Jailbreak" Behavior Assessment Framework]**
Golden Finance reported that on July 3, Anthropic released a proposed framework for assessing the severity of artificial intelligence "jailbreak" behaviors and detailed the cybersecurity measures implemented in the globally launched Claude Fable 5 model. The company categorized cybersecurity-related use cases into four types, ranging from "prohibited use" to "harmless use." Prohibited uses include activities such as ransomware development, malware creation, and cyber-physical infrastructure disruption. Until more robust control mechanisms are established, high-risk activities with "dual-use" attributes, such as penetration testing, will be intercepted.
Anthropic introduced the Cyber Jailbreak Severity (CJS) rating system, which is divided into five levels from CJS-0 to CJS-4, and evaluates severity based on four dimensions. Additionally, Anthropic launched the HackerOne program, inviting security researchers to submit potential model jailbreak cases. (Sina Finance)
Share To
Timeline
HotFlash
APP
X
Telegram
CopyLink