Unverified50% confidenceFactExact time
Decision-1 was the only cloud model in the test that was fooled by a prompt injection command hidden in a comment, misclassifying the customer as happy with a probability of 0.68.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/11/2026
First Seen
Valid until: 1/9/2027
Sources
Related Entities
Related Claims
UnverifiedMoE models only activate a small subset of expert sub-networks during each inference, and Naive-N0.5-Flash has only approximately 5% of its parameters working during actual inference68% similarUnverifiedRanked by question difficulty, persistence rate was 91.4% on the 60 questions that all models answered correctly, but only 73.5% on the 52 questions that at least one model got wrong, indicating that sycophancy tracks uncertainty.68% similarUnverifiedA single "trusted source" prompt caused 7 out of 8 tested models to reverse between 45% and 88% of their correct answers.66% similarUnverifiedBenign paraphrases shifted support judgments by only 0%–2.5% under the blind-verifier prompt.66% similarUnverifiedAdding a "Do not guess" instruction to the prompt can reduce the proportion of fabricated false claims by a model from 71% to 20%.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/998806API
curl https://kongchang.com/api/v1/knowledge/claims/998806MCP
get_claim(id=998806)