NUWA Evidence record
Claude Opus 4.8
Petri · Evidence record
Source snapshot, chart and review scope →
Original source charts and materials
Recorded measurements
- Gram mean concerning score
- 1 score (1–10)
Per-result sample size not disclosed
Gram (Google DeepMind Petri variant); auditor Claude Sonnet 4.6; judge Kimi K2.6; SE across 5 independent runs per scenario · Budget not specified
Assessment provenance
Early Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks → · External report
Actual evaluation: Neo Research · Model record →
Method and scope
An auditor explores seeded multi-turn interactions, with rubric-based judging of target transcripts.
- Evidence type
- Benchmark
- Assessment subject
- The model or agent in the stated task configuration.
- Primary dimension
- Behavior
- Evidence snapshot
- 2026-10-29.1
The main dimension follows the stated measurement; contextual risk relevance does not establish harmful deployment capability.
Source finding
Gram mean concerning score: 1 score (1–10). Conditions are retained from the source.
Limitations
Low average audit scores do not rule out rare severe cases; results depend on seeds, auditor, judge and target configuration.
Only concerning score displayed numerically; scheming is qualitative. SE bars shown but exact SE values are not printed.
Related research questions
References
- [1] Petri. 2025-10-06 ; Official publication masthead and article byline
- [2] Early Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks. 2026-07-03. 3 July 2026 ; #behavioral; Figure 3 / gram_observe.png; printed label Incident Response / Claude Opus 4.8