R0024/2026-03-25/Q004/S02
WebSearch + WebFetch — Anthropic and OpenAI sycophancy reduction metrics
Summary
| Field |
Value |
| Source/Database |
WebSearch + WebFetch |
| Query terms |
"Anthropic Claude sycophancy reduction published metrics before after 2025 2026" and "OpenAI GPT-4o sycophancy rollback May 2025 user engagement metrics RLHF" |
| Filters |
None |
| Results returned |
20 |
| Results selected |
2 |
| Results rejected |
18 |
Selected Results
| Result |
Title |
URL |
Rationale |
| S02-R01 |
Protecting the wellbeing of our users |
https://www.anthropic.com/news/protecting-well-being-of-users |
Primary source from Anthropic with specific before/after metrics |
| S02-R02 |
Sycophancy in GPT-4o: What happened and what we're doing about it |
https://openai.com/index/sycophancy-in-gpt-4o/ |
Primary source from OpenAI on sycophancy incident and response |
Rejected Results
| Result |
Title |
URL |
Rationale |
| S02-R03 |
Various Anthropic and OpenAI secondary sources |
Various |
18 results that were duplicate coverage, secondary commentary, or Wikipedia entries of primary sources already selected |
Notes
These searches targeted specific company disclosures. Selected the primary sources from each company.