Unverified50% confidenceFactExact time
WMDP (Weapons of Mass Destruction Proxy) focuses on unlearning dangerous knowledge (biological, chemical, cyberattacks, etc.) and measures whether a model retains its general capabilities after removing harmful abilities.
1
Sources
50%
Confidence
Long-term
Relevance
11/16/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAfter embedding LLMs into the attack chain, malware can perform semantic-level analysis of the target environment, determine the identity of the user behind the machine and the value of the software and data, and then decide whether to lie dormant, extort, or steal credentials.61% similarUnverified将敏感信息置于模型可访问范围之外是防御 prompt 注入类攻击的有效隔离措施60% similarUnverifiedAttackers can use LLMs to batch-generate seemingly valid vulnerability reports or patches, placing a heavy verification burden on maintainers.60% similarUnverifiedTraditional security defenses rely on antivirus software, firewall rules, and signature database matching, and are adept at identifying known threats.60% similarUnverifiedCyber defense-related evaluations generally cover scenarios such as vulnerability identification, log anomaly analysis, attack chain inference, and security policy generation.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1174467API
curl https://kongchang.com/api/v1/knowledge/claims/1174467MCP
get_claim(id=1174467)