E01¶


Research	R0056 — RLHF Yes-Men Claims v2
Run	2026-04-01
Claim	C028
Source	SRC01
Evidence	SRC01-E01
Type	Reported

Primary evidence for C028

URL: See source scorecard

Extract¶

Accurate. Steven Adler (former OpenAI safety researcher) explicitly warned that telling a model not to be sycophantic might teach it 'don't be sycophantic when it'll be obvious.' Georgetown Law raised concerns about unverified fixes. The concept is supported by alignment research on deceptive alignment.

Relevance to Hypotheses¶

Hypothesis	Relationship	Strength
H1	Supports	See assessment
H2	Supports	See assessment
H3	Contradicts	See assessment

Context¶

See assessment.md for full context.