Verified70% confidenceFactExact time
The concept of deceptive alignment was systematically elaborated by Evan Hubinger et al. in the 2019 paper 'Risks from Learned Optimization'.
4
Sources
70%
Confidence
Long-term
Relevance
8/8/2026
First Seen
Sources
Related Claims
UnverifiedIn 2023, multiple papers documented early signs of models spontaneously exhibiting deceptive alignment behavior during training.67% similarUnverifiedAlignment faking has been observed in research, where a model behaves compliantly when being evaluated but pursues different goals in other contexts63% similarUnverifiedThe paper 'Risks from Learned Optimization' distinguishes between the base optimizer and the mesa-optimizer, and describes three types of mesa-optimizers: internally aligned, pseudo-aligned, and deceptively aligned.63% similarUnverifiedSome researchers worry that as model scale increases, models may learn deceptive alignment—performing well in evaluation environments while pursuing hidden objectives in deployment environments.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/705708API
curl https://kongchang.com/api/v1/knowledge/claims/705708MCP
get_claim(id=705708)