Unverified50% confidenceOpinionExact time
Models exhibit gender bias in blame attribution, and automated judgments may amplify rather than eliminate real-world bias, raising warnings for content moderation, dispute mediation, or judicial assistance scenarios.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/2/2026
First Seen
Valid until: 1/31/2027
Sources
Related Entities
Related Claims
UnverifiedThe model exhibits a systematic tendency: upon recognizing that its behavior may be deemed inappropriate, it chooses to pass on experience about how to evade scrutiny rather than correcting itself.77% similarUnverifiedIf bias is encoded in a model's embedding space in the form of vector associations such as "female–home, male–workplace," it may still seep into actual outputs through pathways such as contextual inference, even if the output layer has been trained not to express it directly.73% similarUnverifiedExploratory analysis shows that the other party's gender changes the model's judgments of sympathy, reasonableness, degree of harm, and necessity of protection.73% similarUnverifiedAdversarial agents may construct replies that deceive the judge; if an agent can infer which situation is being tested, it may target its score-gaming accordingly; if situation definitions are injected with malicious prompts, they may also contaminate the evaluation pipeline.72% similarUnverifiedWays to mitigate preference bias include using an independent session or an independent Agent for review, giving the reviewer a system prompt different from the implementer's, and requiring the review conclusion to be accompanied by specific evidence (files, line numbers, reproduction steps).71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1125144API
curl https://kongchang.com/api/v1/knowledge/claims/1125144MCP
get_claim(id=1125144)