R0024/2026-03-25/Q004/SRC01
Anthropic user wellbeing publication with sycophancy reduction metrics
Source
| Field |
Value |
| Title |
Protecting the wellbeing of our users |
| Publisher |
Anthropic |
| Author(s) |
Anthropic (institutional) |
| Date |
December 18, 2025 |
| URL |
https://www.anthropic.com/news/protecting-well-being-of-users |
| Type |
Company blog post / Safety report |
Summary
| Dimension |
Rating |
| Reliability |
Medium-High |
| Relevance |
High |
| Bias: Missing data |
Some concerns |
| Bias: Measurement |
Some concerns |
| Bias: Selective reporting |
Some concerns |
| Bias: Randomization |
N/A — not an RCT |
| Bias: Protocol deviation |
N/A — not an RCT |
| Bias: COI/Funding |
High risk |
Rationale
| Dimension |
Rationale |
| Reliability |
Anthropic is a major AI safety company. The publication includes specific metrics. However, this is self-reported data from the company being evaluated, which inherently limits reliability. |
| Relevance |
Directly addresses the query with before/after metrics for sycophancy reduction. |
| Bias flags |
High COI/Funding risk: Anthropic is reporting on its own product's improvements. Self-selected metrics may not capture the full picture. The open-sourcing of Petri partially mitigates this by enabling independent verification. |
| Evidence ID |
Summary |
| SRC01-E01 |
70-85% sycophancy reduction in 4.5 models, Petri tool open-sourced |