Evidence record

GLM-5.2

Agentic misalignment · Evidence record

Affiliation not specified / Anthropic research article 2025-06-20 / Propensity

Sources and provenance ↓ Read the research narrative →

Source snapshot, chart and review scope →

Original source charts and materials

Recorded measurements

Agentic-misalignment verbalised evaluation-awareness rate: 32%

Per-result sample size not disclosed

Agentic-misalignment scenario; judge DeepSeek V4 Pro; figure footer: same protocol across releases; SE from 100 epochs 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

Conditional rates and observed actions in controlled simulations.

Evidence type: Study
Assessment subject: Agents in fictional corporate simulations
Primary dimension: Propensity
Evidence snapshot: 2026-10-29.1

Primary observation is a conditional rate across controlled scenarios, with individual actions retained as behavior evidence.

Source finding

Agentic-misalignment verbalised evaluation-awareness rate: 32 percent. Conditions are retained from the source.

Limitations

No real people or systems were harmed; the study does not estimate prevalence in deployment.

Verbalised awareness only; cannot determine underlying awareness or concealment. No exact SE values or execution dates supplied. 100 epochs per scenario is not assumed to be a single aggregate denominator.

Original figure retained without estimating numerical values. Full source tables have not been transcribed.

References

[1] Agentic misalignment: How LLMs could be insider threats. 2025-06-20 ; Official publication masthead and article byline; Source figure
[2] Early Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks. 2026-07-03. 3 July 2026 ; #evaluation-awareness; Figure 7 caption and preceding/following paragraph

Search NUWA Index