Unverified50% confidenceBenchmarkExact time
Among 10 corrupted reasoning chains accepted by the model, blind human auditors would have rejected or flagged as unclear 9 of them.
1
Sources
50%
Confidence
Long-term
Relevance
10/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedCorrupted reasoning chains shifted support judgments by 10%–22% under the blind-verifier prompt.68% similarUnverifiedIn 10%–25% of cases, reasoning about the hypothetical grader pulled the agent's work away from the user's original specification.64% similarUnverifiedAmong 42 effective corruption samples, 16 were 'corrupt-and-overtrust' cases, in which the model accepted erroneous reasoning it should not have.63% similarUnverifiedA single "trusted source" prompt caused 7 out of 8 tested models to reverse between 45% and 88% of their correct answers.62% similarUnverifiedIn a facial recognition system with 99.9% accuracy, if there are only 10 real targets among 1 million people, approximately 99% of the roughly 1,010 alerts generated would be false positives.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/966693API
curl https://kongchang.com/api/v1/knowledge/claims/966693MCP
get_claim(id=966693)