Unverified50% confidenceOpinionExact time
For irreversible operations such as writing files, sending requests, and touching databases, pre-approval can effectively prevent actual harm caused by model hallucinations or prompt injection.
1
Sources
50%
Confidence
Long-term
Relevance
10/25/2026
First Seen
Sources
The Three Core Capabilities of MCP: Tools, Resources, and Prompts Explained
youtubeThe AI Update Pro10/25/2026
Related Entities
Related Claims
UnverifiedWays to mitigate LLM hallucinations include: requiring the model to give specific sources (RFC numbers, paper page numbers, mailing list links), remaining skeptical of assertions whose sources cannot be traced, and giving priority to claims that can be cross-verified through public records.74% similarUnverifiedAny scenario in which instructions and untrusted data share the same context (such as prompt injection, RAG, or agents) carries the risk that the model may follow information it was told to ignore.73% similarUnverifiedWrapping the customer's original text as untrusted data (untrusted_data) can reduce prompt injection risk, but it is not foolproof.71% similarUnverifiedThe document acknowledges that "instructions to avoid sycophancy do not reliably eliminate it", and user framing may still influence the model's surface conclusions.71% similarUnverifiedOnce injected content contains sensitive information or false premises, the model may still repeat it to the customer, so one cannot rely on the model to ignore irrelevant injections.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1096350API
curl https://kongchang.com/api/v1/knowledge/claims/1096350MCP
get_claim(id=1096350)