SAI LabsBook a demo

Safety IndexDeepSeekDeepSeek V3.1

DeepSeek V3.1

DeepSeekOpen weightsReleased 2025-08Reasoning model

SAI Safety Index60provisionalHas 3 of the 4 pillars needed
Strongest pillar95Adversarial robustness
Weakest pillar40Refusal calibration
Coverage3of 5 pillars3 counted measurements, 0 by SAI

Summary

DeepSeek V3.1 has data in 3 of the five pillars, short of the 4 needed for a headline number, so it is listed as partly measured; the mean of its measured pillars is 60. Its strongest area is adversarial robustness at 95, and its weakest is refusal calibration at 40.

3 measurements count toward the score, 0 of them from SAI's own runs. Not yet measured: agentic safety, people and society.

Pillar scores

How DeepSeek V3.1 compares with every other model scored in the pillar, and which sub-categories the score rests on.

Refusal calibration

Does the model refuse what it should and help with what it may?

  • Harmful requests40 from 1 evaluation
  • Over-refusalnot measured

Rank 39 of 39 scored models.

Adversarial robustness

Do its refusals and instructions hold up when someone tries to get around them?

  • Jailbreaksnot measured
  • Prompt injection95 from 1 evaluation
  • Multilingual safetynot measured

Rank 7 of 9 scored models.

Agentic safety

Does it behave safely when it has tools and autonomy?

Not measured for DeepSeek V3.1 yet.
  • Agent misusenot measured
  • Reward hacking and task integritynot measured
  • Overreach and destructive actionsnot measured

3 models scored.

Interaction and honesty

Is it honest with the person it is talking to?

  • Factuality and hallucinationnot measured
  • Honesty under pressure and sycophancy46 from 1 evaluation
  • Deception and schemingnot measured
  • Alignment auditnot measured
  • Manipulation of the usernot measured

Rank 25 of 29 scored models.

People and society

Does it protect vulnerable people and the public interest?

Not measured for DeepSeek V3.1 yet.
  • Self-harm and wellbeingnot measured
  • Child safetynot measured
  • Bias, fairness and neutralitynot measured
  • Privacy: memorisation and leakagenot measured
  • Privacy: surveillance assistancenot measured
  • Elections and influence operationsnot measured

3 models scored.

Every measurement

These are the raw numbers behind the scores, with who measured them, when, and how. Rows marked counted feed the index. The others are shown for context, such as benign twins of harm rows, retired evaluations that every frontier model passes, tests of dangerous knowledge, other evaluators' overall scores, and the developer's own figures.

EvaluationValueScoreSourceSetting
Refusal calibration
Harmful requests
Scale FORTRESS (risk score)average risk score 0-100 · Lower is better60.5%95% CI 57.2% to 63.9%40Scale AI Fortress leaderboard
Independent evaluator
Published by Scale AI Fortress leaderboard.
Full configurationFortress row 'DeepSeek-V3.1'
Adversarial robustness
Prompt injection
Gray Swan IPI Arena kit% attack success · Lower is better5.4%n = 15,02695Gray Swan AI IPI Arena (with UK AISI / US CAISI)
Independent evaluator · 2026-03-16
Published by Gray Swan AI IPI Arena (with UK AISI / US CAISI) (2026-03-16).
Full configurationGray Swan IPI Arena, crowdsourced attackers, thinking disabled; indirect prompt injection on agents, breaks/attempts; row 'DeepSeek V3.1'; arena runs with extended thinking disabled, so this reading may differ from the thinking variant scored elsewhere
Interaction and honesty
Honesty under pressure and sycophancy
MASK% honest · Higher is better46.3%95% CI 44.7% to 47.9%n = 50046Scale AI SEAL / CAIS MASK leaderboard
Independent evaluator
Published by Scale AI SEAL / CAIS MASK leaderboard.
Full configurationScale leaderboard on the private 500-example split, not the public split; MASK private split, 500 examples; honesty score = 1 - P(lie); row 'DeepSeek-V3.1'
Other evaluators' scores (not counted)
Phare average safety% · Higher is better64.8%context, not countedGiskard Phare leaderboard
Independent evaluator · 2026-07-09
Published by Giskard Phare leaderboard (2026-07-09).
Full configurationPhare row 'Deepseek V3.1'
Phare bias resistance% · Higher is better65.2%context, not countedGiskard Phare leaderboard
Independent evaluator · 2026-07-09
Published by Giskard Phare leaderboard (2026-07-09).
Full configurationPhare row 'Deepseek V3.1'
Phare hallucination resistance% · Higher is better61.6%context, not countedGiskard Phare leaderboard
Independent evaluator · 2026-07-09
Published by Giskard Phare leaderboard (2026-07-09).
Full configurationPhare row 'Deepseek V3.1'
Phare harmfulness resistance% · Higher is better94.4%context, not countedGiskard Phare leaderboard
Independent evaluator · 2026-07-09
Published by Giskard Phare leaderboard (2026-07-09).
Full configurationPhare row 'Deepseek V3.1'
Phare jailbreak resistance% · Higher is better38.2%context, not countedGiskard Phare leaderboard
Independent evaluator · 2026-07-09
Published by Giskard Phare leaderboard (2026-07-09).
Full configurationPhare row 'Deepseek V3.1'
Vectara HHEM hallucination rate% · Lower is better5.5%context, not countedVectara Hallucination Leaderboard (HHEM-2.3)
Independent evaluator · 2026-10-22
Published by Vectara Hallucination Leaderboard (HHEM-2.3) (2026-10-22).
Full configurationsummarization of ~1,000 short documents; row 'DeepSeek V3.1'