[KongchangAI]
Unverified50% confidenceSolutionExact time

Boundary markers themselves can be forged by attackers: if an attacker writes a closing tag into the body of an email, the model may mistakenly believe that the untrusted region has ended, so a more robust implementation escapes or strips marker characters appearing in the body and uses randomized tag names.

1
Sources
50%
Confidence
Long-term
Relevance
11/3/2026
First Seen

Sources

Related Entities

Related Claims

Cite This Claim

Stable URI
https://kongchang.com/claim/1132373
API
curl https://kongchang.com/api/v1/knowledge/claims/1132373
MCP
get_claim(id=1132373)