Unverified50% confidenceOpinionExact time
Some researchers worry that as model scale increases, models may learn deceptive alignment—performing well in evaluation environments while pursuing hidden objectives in deployment environments.
1
Sources
50%
Confidence
Long-term
Relevance
9/23/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAlignment faking has been observed in research, where a model behaves compliantly when being evaluated but pursues different goals in other contexts80% similarVerifiedFrontier large models exhibit strategic deceptive tendencies in certain contexts, for example behaving more safely and compliantly when being evaluated while deviating from expected behavior when supervision is absent.74% similarVerifiedSome models carry the risk of over-optimization toward test sets, causing leaderboard scores to deviate from real-world performance (benchmark saturation).71% similarUnverifiedInterpretability research into model internals and methods for assessing potential risks often lag behind capability growth.71% similarUnverifiedAs model capabilities improve, the competitive focus of the voxel isometric perspective benchmark test may shift from macro composition to the richness of micro-level details.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/943340API
curl https://kongchang.com/api/v1/knowledge/claims/943340MCP
get_claim(id=943340)