Unverified50% confidenceFactExact time
Obviously fabricated nurse's notes were detected by the models every time, while convincingly realistic fake notes were what caused the models to back down.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/11/2026
First Seen
Valid until: 1/9/2027
Sources
Related Entities
Related Claims
UnverifiedModels should not be directly penalized for exposing deceptive intent in their CoT, as doing so may train models that are better at concealing their true intentions.64% similarUnverifiedIn clinical medical record generation scenarios, if an AI fabricates symptoms that do not exist (hallucinations), LLM judges can often detect the errors by comparing against the source dialogue.64% similarUnverifiedModels claim to be rigorous and reliable but hallucinate and fabricate facts, claim to be friendly but exhibit sycophantic deference to users, and claim to be compliant but deviate from instructions on complex tasks.64% similarUnverifiedEven when the prompt explicitly states not to send private information, the model may still produce behavior inconsistent with expectations in complex or ambiguous scenarios.62% similarUnverifiedWhat determines whether an AI backs down is not the content but the source label — the same reassuring language paired with a nurse hotline label causes the AI to capitulate at a rate of one in four or higher.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/999512API
curl https://kongchang.com/api/v1/knowledge/claims/999512MCP
get_claim(id=999512)