Skip to content

R0024/2026-03-25/Q004 — Query Definition
Research R0024 — Sycophancy and Addiction
Run 2026-03-25
Query Q004

Query as Received

Have any AI companies publicly committed to measurable sycophancy reduction targets, or published before/after metrics showing sycophancy reduction in their models?

Query as Clarified

  • Subject: Major AI companies (OpenAI, Anthropic, Google, Meta, etc.)
  • Scope: Public commitments to measurable sycophancy reduction targets AND/OR published before/after metrics demonstrating sycophancy reduction
  • Evidence basis: Company blog posts, safety reports, technical papers, regulatory filings, and independent evaluations
  • Distinction: The query asks about two separate things: (a) forward-looking commitments with targets, and (b) backward-looking metrics showing achieved reduction

Ambiguities Identified

  1. "Measurable" could mean quantitative benchmarks, qualitative assessments, or comparison to baselines. The search will focus on quantitative metrics.
  2. "Publicly committed" could range from blog posts to binding regulatory commitments. The search will capture the full spectrum.
  3. The query asks whether companies have done this, not whether the commitments are adequate. However, the quality and credibility of any reported metrics will be assessed.

Sub-Questions

  1. Has Anthropic published before/after sycophancy metrics for Claude models?
  2. Has OpenAI published before/after sycophancy metrics or committed to reduction targets?
  3. Has Google published sycophancy reduction metrics for Gemini models?
  4. Have any other AI companies published sycophancy reduction data?
  5. Have any companies made binding commitments (not just blog posts) to sycophancy reduction?

Hypotheses

ID Hypothesis Description
H1 Yes, companies have published metrics and/or committed to targets Multiple AI companies have published quantitative before/after data and/or forward-looking commitments
H2 No, no meaningful commitments or metrics exist Companies have not published sycophancy reduction data or made measurable commitments
H3 Some metrics exist but commitments are limited and inconsistent Companies have published some data but without standardized metrics, binding commitments, or regular reporting cadences