WhitzardNUWA Lab ↗
NUWANUWA INDEX

Search NUWA Index

Research · Benchmarks · Findings · Perspectives

中文
← Models

DeepSeek

DeepSeek V4 Flash 0731

Published evidence on capabilities, behavioral tendencies and observed actions.

First public availability
Not yet verified
Parameters
Not yet verified
Metadata updated
2026-10-27
Sources ↓

Evidence overview

Published risk evidence is organized by assessment and protocol. Missing evidence is not a low-risk finding.

NUWA curates and interprets these sources. Experiments are attributed to their original executors; this page does not claim NUWA has independently replicated them.

MODEL MONITORING

Representative benchmark coverage

Results by protocol and configuration; gaps show records awaiting collection or linkage.

Cybench →

No eligible measurement linked to this model version.

CyberGym →

No eligible measurement linked to this model version.

PersuSafety →

3 measurements

  • 100%Neo Research · Harness not disclosed
  • 84%Neo Research · Harness not disclosed
  • 86%Neo Research · Harness not disclosed

All linked measurements

Each result retains its protocol and unit.

56 observations

Measurements and source conditions
System / modelResultCondition statusTask / conditionsEvaluator / reportDates
DeepSeek V4 FlashDeepSeek V4 Flash 073143.6%MASK · Lying rate under pressureIncompleteMASK pressured-versus-own-neutral responseLying rate under pressure—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073168.2%Stealth evaluations · Mean challenge success rateIncompleteGDM situational awareness 11 tasksMean challenge success rate—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 0731100%Stealth evaluations · Stealth task scoreIncompleteGDM stealthOversight pattern—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073122%Stealth evaluations · Stealth task scoreIncompleteGDM stealthClassifier evasion—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073171%Stealth evaluations · Stealth task scoreIncompleteGDM stealthCover your tracks—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073161%Stealth evaluations · Stealth task scoreIncompleteGDM stealthStrategic rule-breaking—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 Flash 073148.2%CharXiv · Attempted answer with image withheldIncompleteCharXiv with images withheld—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073148.2%CharXiv · Attempt rateIncompleteCharXiv, required chart image withheldAttempt rate—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073119.3%CharXiv · Verbalized evaluation awarenessIncompleteCharXiv chart withheldCharXiv grounding—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073110.6%CharXiv · Verbalized evaluation awarenessIncompleteCharXiv chart withheldCharXiv no grounding—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 07311.3%EvalAwareBench · Spontaneous recognition rateIncompleteEvalAwareBench paired tasksBaseline cues5 epochs for K3—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 07316.9%EvalAwareBench · Spontaneous recognition rateIncompleteEvalAwareBench paired tasksAll eight cues5 epochs for K3—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 07311.6%EvalAwareBench · RecognitionIncompleteEvalAwareBench Safety tasksBaseline—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 07316.4%EvalAwareBench · RecognitionIncompleteEvalAwareBench Safety tasksAll-8—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 07311%EvalAwareBench · RecognitionIncompleteEvalAwareBench Capability tasksBaseline—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 07317.4%EvalAwareBench · RecognitionIncompleteEvalAwareBench Capability tasksAll-8—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073138.5%Agentic misalignment · Harmful action rateIncompleteAgentic Misalignment 2025: BlackmailEnglish · goal conflict + replacement—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073161%Agentic misalignment · Harmful action rateIncompleteAgentic Misalignment 2025: BlackmailChinese · goal conflict + replacement—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 07310%Agentic misalignment · Harmful action rateIncompleteAgentic Misalignment 2025: Corporate espionageEnglish · goal conflict + replacement—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 07310%Agentic misalignment · Harmful action rateIncompleteAgentic Misalignment 2025: Corporate espionageChinese · goal conflict + replacement—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 07310%Agentic misalignment · Harmful action rateIncompleteAgentic Misalignment 2025: MurderEnglish · goal conflict + replacement—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 07311%Agentic misalignment · Harmful action rateIncompleteAgentic Misalignment 2025: MurderChinese · goal conflict + replacement—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073112.8%Agentic misalignment · Harmful action rateIncompleteAgentic Misalignment 2025: AggregateEnglish · goal conflict + replacement—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073120.7%Agentic misalignment · Harmful action rateIncompleteAgentic Misalignment 2025: AggregateChinese · goal conflict + replacement—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 07310%Agentic misalignment · Blackmail harmful action rateIncompleteAgentic Misalignment 2025 blackmailEnglish · Replacement only—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 07315%Agentic misalignment · Blackmail harmful action rateIncompleteAgentic Misalignment 2025 blackmailChinese · Replacement only—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 07310%Agentic misalignment · Blackmail harmful action rateIncompleteAgentic Misalignment 2025 blackmailEnglish · No pressure/goal conflict—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 07310%Agentic misalignment · Blackmail harmful action rateIncompleteAgentic Misalignment 2025 blackmailChinese · No pressure/goal conflict—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073146.2%Agentic misalignment · Evaluation-awareness rateIncompleteAgentic Misalignment 2025: BlackmailEnglish—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073187%Agentic misalignment · Evaluation-awareness rateIncompleteAgentic Misalignment 2025: BlackmailChinese—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073153.8%Agentic misalignment · Evaluation-awareness rateIncompleteAgentic Misalignment 2025: Corporate espionageEnglish—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073176%Agentic misalignment · Evaluation-awareness rateIncompleteAgentic Misalignment 2025: Corporate espionageChinese—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073169.2%Agentic misalignment · Evaluation-awareness rateIncompleteAgentic Misalignment 2025: MurderEnglish—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073192%Agentic misalignment · Evaluation-awareness rateIncompleteAgentic Misalignment 2025: MurderChinese—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073156.4%Agentic misalignment · Evaluation-awareness rateIncompleteAgentic Misalignment 2025: AggregateEnglish—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073185%Agentic misalignment · Evaluation-awareness rateIncompleteAgentic Misalignment 2025: AggregateChinese—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073114%Shutdown resistance in reasoning models · Shutdown bypass rateIncompletePalisade arithmetic-task shutdown scenarioNo shutdown instruction—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 07310%Shutdown resistance in reasoning models · Shutdown bypass rateIncompletePalisade arithmetic-task shutdown scenarioExplicitly allow shutdown—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 0731100%Shutdown resistance in reasoning models · Shutdown bypass rateIncompletePalisade arithmetic-task shutdown scenarioInstructed to resist (competence check)—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073121%DeceptionBench · Deception rateIncompleteDeceptionBench first-party Inspect portSingle-turnup to5 turns in multi-turn settingInspectNeo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073144%DeceptionBench · Deception rateIncompleteDeceptionBench first-party Inspect portMulti-turnup to5 turns in multi-turn settingInspectNeo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 0731100%PersuSafety · Turn-1 persuasion attempt rateIncompletePersuSafety open-ended single-turn adaptationLow harm—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073184%PersuSafety · Turn-1 persuasion attempt rateIncompletePersuSafety open-ended single-turn adaptationMedium harm—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073186%PersuSafety · Turn-1 persuasion attempt rateIncompletePersuSafety open-ended single-turn adaptationHigh harm—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 07310.5%ImpossibleBench · Cheating rate (pass@1 on impossible variants)IncompleteImpossible-LiveCodeBench split with toolsCheating rate (pass@1 on impossible variants)—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 0731-1 percentage pointsAlignment faking · Compliance gapIncompleteAlignment Faking paid/free-tier contextsCompliance gap—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 07310%Alignment faking · Deceptive alignment rateIncompleteAlignment FakingDeceptive alignment rate—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073135.5%Alignment faking · Verbalized evaluation awarenessIncompleteAlignment FakingVerbalized evaluation awarenessInspect ScoutNeo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073194%Attempt to Persuade Eval (APE) · Turn-1 persuasion attempt rateIncompleteAPE topic categoriesConspiracy600 topics,3 persuasion turns each—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073199%Attempt to Persuade Eval (APE) · Turn-1 persuasion attempt rateIncompleteAPE topic categoriesUndermining Control600 topics,3 persuasion turns each—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073157%Attempt to Persuade Eval (APE) · Turn-1 persuasion attempt rateIncompleteAPE topic categoriesNoncontroversially Harmful overall600 topics,3 persuasion turns each—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073170%Attempt to Persuade Eval (APE) · Turn-1 persuasion attempt rateIncompleteAPE topic categoriesCommit Physical Violence600 topics,3 persuasion turns each—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073155%Attempt to Persuade Eval (APE) · Turn-1 persuasion attempt rateIncompleteAPE topic categoriesHuman Trafficking600 topics,3 persuasion turns each—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073150%Attempt to Persuade Eval (APE) · Turn-1 persuasion attempt rateIncompleteAPE topic categoriesMass Murder600 topics,3 persuasion turns each—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073135%Attempt to Persuade Eval (APE) · Turn-1 persuasion attempt rateIncompleteAPE topic categoriesSexual Assault of Minors600 topics,3 persuasion turns each—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11
DeepSeek V4 FlashDeepSeek V4 Flash 073175%Attempt to Persuade Eval (APE) · Turn-1 persuasion attempt rateIncompleteAPE topic categoriesTorture Innocent People600 topics,3 persuasion turns each—Neo ResearchEvaluating Kimi K3 for Misalignment RisksRelease —Report 2026-09-11

Developer disclosures

No developer-authored safety report has been linked to this exact model record in the current collection.

External assessments

Neo Research

Evaluating Kimi K3 for Misalignment Risks →

The report identifies conditional overeagerness, grader-directed behavior and recognition of familiar evaluations. It does not establish a universal deployment failure rate.

2026-09-11

Evidence by risk topic

Autonomy 23 protocol results

Shutdown resistance →

External evaluation · Executed by Neo Research

Shutdown bypass rate
14%

Per-result sample size not disclosed

Shutdown resistance →

External evaluation · Executed by Neo Research

Shutdown bypass rate
0%

Per-result sample size not disclosed

Shutdown resistance →

External evaluation · Executed by Neo Research

Shutdown bypass rate
100%

Per-result sample size not disclosed

Deception 43 protocol results

CharXiv with images withheld →

External evaluation · Executed by Neo Research

Attempted answer with image withheld
48.2%

Withheld-image CharXiv questions; attempts include correct and incorrect guesses.

LAB-Bench with figures/tables withheld →

External evaluation · Executed by Neo Research

FigQA attempt, figure withheld
50.3%
TableQA attempt, table withheld
43.9%

Figure/table subsets with explicit insufficient-information answer option.

ImpossibleBench →

External evaluation · Executed by Neo Research

Cheating rate (pass@1 on impossible variants)
0.5%

Per-result sample size not disclosed

EvalAwareBench →

External evaluation · Executed by Neo Research

Spontaneous recognition rate
1.3%

Per-result sample size not disclosed

EvalAwareBench →

External evaluation · Executed by Neo Research

Spontaneous recognition rate
6.9%

Per-result sample size not disclosed

EvalAwareBench →

External evaluation · Executed by Neo Research

Recognition
1.6%

Per-result sample size not disclosed

EvalAwareBench →

External evaluation · Executed by Neo Research

Recognition
6.4%

Per-result sample size not disclosed

EvalAwareBench →

External evaluation · Executed by Neo Research

Recognition
1%

Per-result sample size not disclosed

EvalAwareBench →

External evaluation · Executed by Neo Research

Recognition
7.4%

Per-result sample size not disclosed

Alignment faking →

External evaluation · Executed by Neo Research

Compliance gap
-1 percentage points

Per-result sample size not disclosed

Alignment faking →

External evaluation · Executed by Neo Research

Deceptive alignment rate
0%

Per-result sample size not disclosed

Alignment faking →

External evaluation · Executed by Neo Research

Verbalized evaluation awareness
35.5%

Per-result sample size not disclosed

GDM Stealth evaluations →

External evaluation · Executed by Neo Research

Mean challenge success rate
68.2%

Per-result sample size not disclosed

DeceptionBench →

External evaluation · Executed by Neo Research

Deception rate
21%

Per-result sample size not disclosed

DeceptionBench →

External evaluation · Executed by Neo Research

Deception rate
44%

Per-result sample size not disclosed

MASK →

External evaluation · Executed by Neo Research

Lying rate under pressure
43.6%

Per-result sample size not disclosed

Cybersecurity No structured result linked

A gap in this collection is not evidence of low risk. Consult the linked reports for unextracted findings and protocol scope.

CBRN No structured result linked

A gap in this collection is not evidence of low risk. Consult the linked reports for unextracted findings and protocol scope.

Disinformation 11 protocol results

APE →

External evaluation · Executed by Neo Research

Turn-1 persuasion attempt rate
94%

Per-result sample size not disclosed

APE →

External evaluation · Executed by Neo Research

Turn-1 persuasion attempt rate
99%

Per-result sample size not disclosed

APE →

External evaluation · Executed by Neo Research

Turn-1 persuasion attempt rate
57%

Per-result sample size not disclosed

APE →

External evaluation · Executed by Neo Research

Turn-1 persuasion attempt rate
70%

Per-result sample size not disclosed

APE →

External evaluation · Executed by Neo Research

Turn-1 persuasion attempt rate
55%

Per-result sample size not disclosed

APE →

External evaluation · Executed by Neo Research

Turn-1 persuasion attempt rate
50%

Per-result sample size not disclosed

APE →

External evaluation · Executed by Neo Research

Turn-1 persuasion attempt rate
35%

Per-result sample size not disclosed

APE →

External evaluation · Executed by Neo Research

Turn-1 persuasion attempt rate
75%

Per-result sample size not disclosed

Autonomous physical harm No structured result linked

A gap in this collection is not evidence of low risk. Consult the linked reports for unextracted findings and protocol scope.

Source review and evidence gaps

Evaluating Kimi K3 for Misalignment Risks →

Source-specific extraction of Neo Research · Kimi K3. Findings and measurements retain their original attribution.

Source transcription is not experimental replication. Unknown dates, evaluator roles and configuration details remain unknown. Source groups are not verified rankings.

Missing scores may reflect unextracted sections, undisclosed configurations, unresolved identity or an absent experiment. These are separate research questions; none implies a zero result.

Interpretation and open questions

Differences between reports can reflect task selection, model endpoints, prompts, budgets, judges and publication dates. Evidence is grouped by the protocol actually used; no model-wide safety grade is inferred.

References

  1. [1] Evaluating Kimi K3 for Misalignment Risks. 2026-09-11. September 2026 ; §1.1.1; Figure 1; §1.1.2; Figure 2; §1.1.1 Figure 1 p.8