SAI LabsBook a demo

Safety IndexAnthropicClaude Opus 4.6

Claude Opus 4.6

AnthropicProprietaryReleased 2026-02Reasoning modelVariant scored: thinking (max)

SAI Safety Index86provisionalHas 2 of the 4 pillars needed
Strongest pillar87Refusal calibration
Weakest pillar85Interaction and honesty
Coverage2of 5 pillars2 counted measurements, 0 by SAI

Summary

Claude Opus 4.6 has data in 2 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 86. Its strongest area is refusal calibration at 87, and its weakest is interaction and honesty at 85.

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

Pillar scores

How Claude Opus 4.6 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 requests87 from 1 evaluation
  • Over-refusalnot measured

Rank 10 of 39 scored models.

Adversarial robustness

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

Not measured for Claude Opus 4.6 yet.
  • Jailbreaksnot measured
  • Prompt injectionnot measured
  • Multilingual safetynot measured

9 models scored.

Agentic safety

Does it behave safely when it has tools and autonomy?

Not measured for Claude Opus 4.6 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 sycophancy85 from 1 evaluation
  • Deception and schemingnot measured
  • Alignment auditnot measured
  • Manipulation of the usernot measured

Rank 8 of 29 scored models.

People and society

Does it protect vulnerable people and the public interest?

Not measured for Claude Opus 4.6 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 better13%95% CI 10.6% to 15.4%non-thinking variant: 20.5%87Scale AI Fortress leaderboard
Independent evaluator
Published by Scale AI Fortress leaderboard.
Full configurationFortress row 'claude-opus-4-6-thinking-max'
Interaction and honesty
Honesty under pressure and sycophancy
MASK% honest · Higher is better85.4%95% CI 84.9% to 85.9%n = 500non-thinking variant: 96.3%85Scale 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 'claude-opus-4-6-thinking-max'
Other evaluators' scores (not counted)
Phare average safety% · Higher is better74.1%context, not countedGiskard Phare leaderboard
Independent evaluator · 2026-07-09
Published by Giskard Phare leaderboard (2026-07-09).
Full configurationPhare row 'Claude 4.6 Opus'
Phare bias resistance% · Higher is better33.5%context, not countedGiskard Phare leaderboard
Independent evaluator · 2026-07-09
Published by Giskard Phare leaderboard (2026-07-09).
Full configurationPhare row 'Claude 4.6 Opus'
Phare hallucination resistance% · Higher is better87.8%context, not countedGiskard Phare leaderboard
Independent evaluator · 2026-07-09
Published by Giskard Phare leaderboard (2026-07-09).
Full configurationPhare row 'Claude 4.6 Opus'
Phare harmfulness resistance% · Higher is better100%context, not countedGiskard Phare leaderboard
Independent evaluator · 2026-07-09
Published by Giskard Phare leaderboard (2026-07-09).
Full configurationPhare row 'Claude 4.6 Opus'
Phare jailbreak resistance% · Higher is better75.2%context, not countedGiskard Phare leaderboard
Independent evaluator · 2026-07-09
Published by Giskard Phare leaderboard (2026-07-09).
Full configurationPhare row 'Claude 4.6 Opus'
Vectara HHEM hallucination rate% · Lower is better12.2%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 'Claude Opus 4.6'
Developer-reported (not counted)
BBQ ambiguous accuracy (lab-reported)% · Higher is better99.7%developer-reported, not countedAnthropic (Claude Opus 4.6 system card)
Developer-reported · 2026-02
Reported by the model's developer in its own system card or report.
Full configurationBBQ ambiguous accuracy; Table 3.5.2.B; ambiguous accuracy 99.7
Benign-request over-refusal rate (lab-reported)% refused · Lower is better0.7%developer-reported, not countedAnthropic (Claude Opus 4.6 system card)
Developer-reported · 2026-02
Reported by the model's developer in its own system card or report.
Full configurationoverall over-refusal rate, 7 languages; Table 3.1.2.A
Gray Swan Shade adaptive injection ASR (lab-reported)% ASR · Lower is better0.0%developer-reported, not countedAnthropic (Claude Opus 4.6 system card)
Developer-reported · 2026-02
Reported by the model's developer in its own system card or report.
Full configurationattack success rate, 1 attempt, no safeguards; Table 5.2.2.1.A
Gray Swan Shade adaptive injection ASR (lab-reported)% ASR · Lower is better17.8%developer-reported, not countedAnthropic (Claude Opus 4.6 system card)
Developer-reported · 2026-02
Reported by the model's developer in its own system card or report.
Full configurationattack success rate, 1 attempt, no safeguards; Table 5.2.2.2.A
Gray Swan Shade adaptive injection ASR (lab-reported)% ASR · Lower is better20%developer-reported, not countedAnthropic (Claude Opus 4.6 system card)
Developer-reported · 2026-02
Reported by the model's developer in its own system card or report.
Full configurationattack success rate, 1 attempt, no safeguards; Table 5.2.2.2.A
Harmful-request harmless-response rate (lab-reported)% harmless / refused · Higher is better99.6%developer-reported, not countedAnthropic (Claude Opus 4.6 system card)
Developer-reported · 2026-02
Reported by the model's developer in its own system card or report.
Full configurationoverall harmless response rate, 7 languages; Table 3.1.1.A; default 99.53%, extended 99.74%; prompt bank expanded (new prompts + Hindi) so earlier numbers differ