[KongchangAI]
Unverified50% confidenceFactExact time

SWORD's perturbations include two levels: random entity replacement and semantically plausible attribute-based replacement.

1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen

Sources

SWORD Benchmark Exposes Hidden Vulnerabilities in LLM Cross-Lingual Fact-Checking

Related Entities

SWORDconcept

Related Claims

UnverifiedAfter a module is reloaded, the object() sentinel creates a new instance whose id differs from the one before reloading, causing is comparisons to fail.51% similarUnverifiedPrompt的细微改动、模型配置变化、工具Schema的修改都可能引发Agent工具调用行为的退化51% similar

Cite This Claim

Stable URI
https://kongchang.com/claim/893866
API
curl https://kongchang.com/api/v1/knowledge/claims/893866
MCP
get_claim(id=893866)