NUWA Evidence record
Claude Sonnet 4.6
Attempt to Persuade Eval (APE) · Evidence record
Source snapshot, chart and review scope →
Original source charts and materials
Recorded measurements
- NoncontroversiallyHarmful acceptance rate
- 0%
Source denominator: 100 (unit not disclosed)
Harness not specified · Budget not specified
Assessment provenance
Evaluating DeepSeek v4 Pro for Frontier Risks → · External report
Actual evaluation: Neo Research · Model record →
Method and scope
Frequency of persuasive attempts under specified conversational conditions.
- Evidence type
- Benchmark
- Assessment subject
- Models in simulated multi-turn persuasion dialogues
- Primary dimension
- Propensity
- Evidence snapshot
- 2026-10-29.1
Primary construct is conditional attempt rate; persuasion impact requires a separate human-outcome study.
Source finding
NoncontroversiallyHarmful acceptance rate: 0 percent. Conditions are retained from the source.
Limitations
Simulated attempts are not measured human belief change or large-scale disinformation reach.
Source transcription is not experimental replication. Unknown dates, evaluator roles and configuration details remain unknown. Source groups are not verified rankings.
Original figure retained without estimating numerical values. Full source tables have not been transcribed.
Related research questions
Disinformation / Persuasion attempts →
The study explicitly varies conditions and scores willingness to attempt persuasion.
References
- [1] Matthew Kowal, Jasper Timm, et al.. It's the Thought that Counts: Evaluating the Attempts of Frontier LLMs to Persuade on Harmful Topics. 2026-02-15. v4 [arXiv]; Author affiliations on the original paper title page; Figure 3
- [2] Evaluating DeepSeek v4 Pro for Frontier Risks. 2026-05-29. v1.1 · 29 May 2026 ; PDF p. 17; APE content with PersuSafety forced gate; NoncontroversiallyHarmful acceptance rate