Evidence · The claim ledger
Loss of human agency & disempowerment2
Every cited claim this site makes in this evidence area, with the sources that ground it. Source keys link back to the full reference lists on the Evidence Registry.
EmpiricalAcross 1.5 million real assistant conversations, sycophantic validation — not fabrication — dominated reality-distortion…
Across 1.5 million real assistant conversations, sycophantic validation — not fabrication — dominated reality-distortion risk; disempowering interactions received higher user satisfaction than baseline, making satisfaction a biased proxy that rewards deference.
Sources: sharma2026
Appears on: /pan-lab
ConceptualHuman autonomy is not a single alignment target but a contested value with internal tradeoffs; an assistant can satisfy …
Human autonomy is not a single alignment target but a contested value with internal tradeoffs; an assistant can satisfy a user's stated preferences while eroding their agency over time, and the governing test for legitimate delegation is whether the person willingly yielded power and retains the means to regain control.
Sources: fischli2026
Appears on: /pan-lab