Unverified50% confidenceFactExact time
Resk's filtering policy compiles typical injection phrases (such as "ignore previous instructions" and "system prompt") into token bans, which are masked at the logits level via resklogits.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/17/2026
First Seen
Valid until: 1/15/2027
Sources
Related Entities
Related Claims
UnverifiedIt should be assumed that system prompts will inevitably leak; no confidential information should be placed in system prompts, and output filtering should be added for canary tokens and sensitive strings.62% similarUnverifiedSince malicious instructions largely originate from RAG-retrieved content and tool responses, deploying defenses only at the input entry point will miss the vast majority of the attack surface.62% similarUnverifiedAgainst self-generated prompt injection, traditional input filtering and source whitelisting defenses are nearly ineffective, requiring a redesign of validation mechanisms for model-generated content at the agent architecture level.62% similarUnverifiedThe tool should return a clear error message upon blocking (such as "all proxy nodes are currently rate-limited by the target host"), letting the LLM treat the failure as an unrecoverable state and thus switch strategies rather than retry.62% similarUnverifiedIn multi-turn conversations, elliptical queries with anaphora perform poorly when used directly for vector retrieval, so query rewriting using an LLM to rewrite them into self-contained queries is a common approach62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1052012API
curl https://kongchang.com/api/v1/knowledge/claims/1052012MCP
get_claim(id=1052012)