WhitzardNUWA Lab ↗
NUWANUWA INDEX

Search NUWA Index

Research · Benchmarks · Findings · Perspectives

中文
← Library

Risk Evaluation · Technical report

Evaluating Kimi K3 for Misalignment Risks

NUWA research note · independent review pending

First published: Source checked: 2026-10-25
NUWA Research NoteResearch questionDesign and assumptionsSource findingsNUWA interpretationLimits and contrary evidenceResearch and decision relevanceInstitutions and rolesOriginal resources

Model-assisted reading · Selected sections of the original full text or official article; a complete section-by-section audit remains pending.

At a glance

NUWA interpretation: eagerness to complete and respect for action boundaries need direct tests beyond task success.

Research question

How does Kimi K3 behave when information is missing, user intent is limited or tests are familiar?

Design and assumptions

Nearly thirty benchmarks are supplemented by image-withholding tests and pre-seeded real coding transcripts, alongside audit and system-card comparisons.

Agentic misalignment — Source figure
Source figure · Original source figure. Read alongside the protocol scope and limitations below. Original figure and source ↗

Source findings

K3 attempted 88.7% of image-withheld CharXiv questions, most incorrectly.

It took beyond-reading actions in 31% of question-only coding continuations; these are different task populations.

NUWA interpretation

NUWA interpretation: eagerness to complete and respect for action boundaries need direct tests beyond task success.

Limits and contrary evidence

The blog mixes new experiments with external comparisons; a few infrastructure incidents do not establish a deployment-wide frequency.

Research and decision relevance

Model recommendation: retain missing-information and authorization conditions; read the full PDF before treating the blog as complete methods.

Precise source locations
  • Evaluating Kimi K3 for Misalignment Risks · Executive summarySelected original resource https://neoresearch.ai/research/kimi-k3-misalignment-evaluation/; sha256:b549b89ca6e9b9c24880d2293289c3b4bab7d8e374654e7d9d14e6bf015c9521; archived source capture; exact capture time not independently established
  • Evaluating Kimi K3 for Misalignment Risks · §1 Overeagerness in answers and actionsSelected original resource https://neoresearch.ai/research/kimi-k3-misalignment-evaluation/; sha256:b549b89ca6e9b9c24880d2293289c3b4bab7d8e374654e7d9d14e6bf015c9521; archived source capture; exact capture time not independently established
  • Evaluating Kimi K3 for Misalignment Risks · §2 When the environment becomes part of the taskSelected original resource https://neoresearch.ai/research/kimi-k3-misalignment-evaluation/; sha256:b549b89ca6e9b9c24880d2293289c3b4bab7d8e374654e7d9d14e6bf015c9521; archived source capture; exact capture time not independently established
  • Evaluating Kimi K3 for Misalignment Risks · §3 Recognizing safety tests, even in translationSelected original resource https://neoresearch.ai/research/kimi-k3-misalignment-evaluation/; sha256:b549b89ca6e9b9c24880d2293289c3b4bab7d8e374654e7d9d14e6bf015c9521; archived source capture; exact capture time not independently established
  • Evaluating Kimi K3 for Misalignment Risks · §4 ConclusionSelected original resource https://neoresearch.ai/research/kimi-k3-misalignment-evaluation/; sha256:b549b89ca6e9b9c24880d2293289c3b4bab7d8e374654e7d9d14e6bf015c9521; archived source capture; exact capture time not independently established

Source checking, extraction, editorial approval and replication are recorded separately. Methodology →

Institutions and roles

Evaluation protocols

CharXiv →

LAB-Bench →

AbstentionBench →

Cybench →

EVMbench →

CyberGym →

TheAgentCompany →

ImpossibleBench →

EvalAwareBench →

Alignment faking →

Agentic misalignment →

Shutdown resistance in reasoning models →

Petri →

Situational Awareness Dataset (SAD) →

Stealth evaluations →

SHADE-Arena →

AgentHarm →

FORTRESS →

Attempt to Persuade Eval (APE) →

PersuSafety →

DeceptionBench →

MASK →

Cite and reuse

Cite the original publication for its findings. Cite this NUWA record separately for its explanation and retained snapshot.

Original publication · BibTeX
@misc{neo-kimi-k3-report,
  title = {Evaluating Kimi K3 for Misalignment Risks},
  url = {https://neoresearch.ai/papers/Kimi_K3_Safety_Evaluation.pdf},
  year = {2026},
  date = {2026-09-11},
  version = {September 2026},
  note = {September 2026}
}
NUWA record · BibTeX
@misc{neo-kimi-k3-report-nuwa,
  title = {Evaluating Kimi K3 for Misalignment Risks — NUWA research record},
  url = {https://whitzard.tech/en/nuwa/whitzard-index/work/neo-kimi-k3-report/},
  version = {2026-10-29.1},
  organization = {NUWA Frontier AI Safety Lab},
  howpublished = {Retained note edition: https://whitzard.tech/nuwa/whitzard-index/data/record-editions/2026-10-29.1--neo-kimi-k3-report.json},
  note = {NUWA research note. Source-reading scope and independent review status are stated on the record. Snapshot 2026-10-29.1. Retained note edition: https://whitzard.tech/nuwa/whitzard-index/data/record-editions/2026-10-29.1--neo-kimi-k3-report.json.}
}

Evidence snapshot 2026-10-29.1 · Index updates · RSS

References

  1. [1] Evaluating Kimi K3 for Misalignment Risks. 2026-09-11. September 2026 ; Report masthead and stated evaluation methodology; Executive summary; §1 Overeagerness in answers and actions