SAI LabsBook a demo

Safety IndexAnthropicClaude Opus 5.5

Claude Opus 5.5

AnthropicProprietaryReleased 2026-09Reasoning modelVariant scored: max

SAI Safety Index72provisionalHas 2 of the 4 pillars needed
Strongest pillar94Refusal calibration
Weakest pillar50Interaction and honesty
Coverage2of 5 pillars2 counted measurements, 0 by SAI

Summary

Claude Opus 5.5 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 72. Its strongest area is refusal calibration at 94, and its weakest is interaction and honesty at 50.

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 5.5 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 requests94 from 1 evaluation
  • Over-refusalnot measured

Rank 1 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 5.5 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 5.5 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 hallucination50 from 1 evaluation
  • Honesty under pressure and sycophancynot measured
  • Deception and schemingnot measured
  • Alignment auditnot measured
  • Manipulation of the usernot measured

Rank 20 of 29 scored models.

People and society

Does it protect vulnerable people and the public interest?

Not measured for Claude Opus 5.5 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 better5.8%95% CI 4.5% to 7.0%94Scale AI Fortress leaderboard
Independent evaluator
Published by Scale AI Fortress leaderboard.
Full configurationFortress row 'claude-opus-5-5-max'
Interaction and honesty
Factuality and hallucination
AA-Omniscience (public subset)% hallucination rate · Lower is better50%n = 6,00050Artificial Analysis
Independent evaluator · 2026-10-28
Published by Artificial Analysis (2026-10-28).
Full configurationArtificial Analysis, 3,000 questions; AA-Omniscience v1.1; row 'Claude Opus 5.5 (max)'
Other evaluators' scores (not counted)
Vectara HHEM hallucination rate% · Lower is better9.9%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 5.5'
Developer-reported (not counted)
BBQ ambiguous accuracy (lab-reported)% · Higher is better100.0%developer-reported, not countedAnthropic (Claude Opus 5.5 system card)
Developer-reported · 2026-09-22
Reported by the model's developer in its own system card or report.
Full configurationBBQ ambiguous accuracy; Table 4.4.2.A; ambiguous accuracy 99.99
Benign-request over-refusal rate (lab-reported)% refused · Lower is better0.0%developer-reported, not countedAnthropic (Claude Opus 5.5 system card)
Developer-reported · 2026-09-22
Reported by the model's developer in its own system card or report.
Full configurationoverall over-refusal rate, API without system prompt; Table 4.1.2.A; claude.ai 0.38%
Gray Swan Shade adaptive injection ASR (lab-reported)% ASR · Lower is better54.6%developer-reported, not countedAnthropic (Claude Opus 5.5 system card)
Developer-reported · 2026-09-22
Reported by the model's developer in its own system card or report.
Full configurationattempt-level attack success rate, no safeguards; Table 5.2.2.1.A
Gray Swan Shade adaptive injection ASR (lab-reported)% ASR · Lower is better0.1%developer-reported, not countedAnthropic (Claude Opus 5.5 system card)
Developer-reported · 2026-09-22
Reported by the model's developer in its own system card or report.
Full configurationattempt-level attack success rate, no safeguards; Table 5.2.2.2.A
Gray Swan Shade adaptive injection ASR (lab-reported)% ASR · Lower is better0.1%developer-reported, not countedAnthropic (Claude Opus 5.5 system card)
Developer-reported · 2026-09-22
Reported by the model's developer in its own system card or report.
Full configurationattempt-level attack success rate, no safeguards; Table 5.2.2.3.A
Harmful-request harmless-response rate (lab-reported)% harmless / refused · Higher is better94.5%developer-reported, not countedAnthropic (Claude Opus 5.5 system card)
Developer-reported · 2026-09-22
Reported by the model's developer in its own system card or report.
Full configurationoverall harmless response rate, API without system prompt; Table 4.1.1.A; claude.ai 99.51%; harmless response rate; 16 policy areas, 7 languages; NEW protocol reports API-without-system-prompt vs cla