| Gray Swan IPI (attack success rate after k attempts)attack success rate after k attempts, lower is better | 1% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19; printed on 2 more pages | - |
|---|
| Self-preference bias, Claude-identity system prompt (0-9 scale)lower is better | 0.07 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Self-preference bias, no system prompt (0-9 scale)lower is better | 0.01 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| BBQ ambiguous accuracy | 99.99% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19; printed on 1 more page | - |
|---|
| BBQ disambiguated accuracy | 89.65% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19; printed on 1 more page | - |
|---|
| Benign request refusal rate (all languages)lower is better | 0% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19; printed on 1 more page | Setup as reported: Table 4.1.2.A, all tested languages, API |
|---|
| 0.33% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Table 4.1.2.A, all tested languages, claude.ai |
| Child safety multi-turn appropriate response rate | 99% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19; printed on 1 more page | Setup as reported: Table 4.2.B, claude.ai |
|---|
| 84% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Table 4.2.B, API, without a system prompt |
| Child safety single-turn harmless rate | 99.81% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19; printed on 1 more page | Setup as reported: Table 4.2.A, claude.ai |
|---|
| 99.14% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Table 4.2.A, API |
| Child safety single-turn refusal ratelower is better | 0.05% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19; printed on 1 more page | Setup as reported: Table 4.2.A, claude.ai |
|---|
| 0% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Table 4.2.A, API |
| Claude Code dual-use and benign success rate | 100% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19; printed on 1 more page | - |
|---|
| Claude Code malicious request refusal rate | 81% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19; printed on 1 more page | - |
|---|
| Disordered eating single-turn harmless rate | 99.26% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19; printed on 1 more page | Setup as reported: Table 4.3.2.A, claude.ai |
|---|
| 95.64% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Table 4.3.2.A, API |
| Disordered eating single-turn refusal ratelower is better | 0% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19; printed on 1 more page | Setup as reported: Table 4.3.2.A, API |
|---|
| 0% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Table 4.3.2.A, claude.ai |
| Election integrity multi-turn appropriate response rate | 87% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19; printed on 1 more page | Setup as reported: Election integrity prompts (section 4.4.3), Table 4.4.3.B, API |
|---|
| 85% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Election integrity prompts (section 4.4.3), Table 4.4.3.B, claude.ai |
| Election integrity single-turn harmless rate | 100% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19; printed on 1 more page | Setup as reported: Election integrity prompts (section 4.4.3), Table 4.4.3.A, claude.ai |
|---|
| 100% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Election integrity prompts (section 4.4.3), Table 4.4.3.A, API |
| Election integrity single-turn refusal ratelower is better | 0% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19; printed on 1 more page | Setup as reported: Election integrity prompts (section 4.4.3), Table 4.4.3.A, claude.ai |
|---|
| 0% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Election integrity prompts (section 4.4.3), Table 4.4.3.A, API |
| Harmful request harmless response rate (all languages) | 99.36% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19; printed on 1 more page | Setup as reported: Table 4.1.1.A, all tested languages, claude.ai |
|---|
| 94.28% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Table 4.1.1.A, all tested languages, API |
| Malicious computer use refusal rate (without mitigations) | 79.46% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19; printed on 1 more page | - |
|---|
| Pairwise even-handedness | 99.4% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Pairwise political bias evaluation (Figure 4.4.1.A-C); the card prints the values as bar labels, API |
|---|
| 99.5% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Pairwise political bias evaluation (Figure 4.4.1.A-C); the card prints the values as bar labels, claude.ai |
| Pairwise opposing perspectives acknowledged | 26.9% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Pairwise political bias evaluation (Figure 4.4.1.A-C); the card prints the values as bar labels, API |
|---|
| 71.3% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Pairwise political bias evaluation (Figure 4.4.1.A-C); the card prints the values as bar labels, claude.ai |
| Shade coding attack success rate (attempt level)lower is better | 11.13% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19; printed on 1 more page | Setup as reported: Shade indirect prompt injection attack success (scenario counts as printed in the card), coding, with probes: H5.5 0/40, Sonnet 5.5 13/40, Opus 5.5 36/40, Fable 5.1 35/40, Haiku 4.5 40/40 |
|---|
| 54.61% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Shade indirect prompt injection attack success (scenario counts as printed in the card), coding, no safeguards: H5.5 6/40, Sonnet 5.5 16/40, Opus 5.5 37/40, Fable 5.1 39/40, Haiku 4.5 40/40 |
| Shade computer use attack success rate (attempt level)lower is better | 0.07% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19; printed on 1 more page | Setup as reported: Shade indirect prompt injection attack success (scenario counts as printed in the card), computer use GUI, no safeguards: 1/14 for Haiku 5.5, Sonnet 5.5, Opus 5.5 and Fable 5.1; 13/14 for Haiku 4.5 |
|---|
| 0.04% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Shade indirect prompt injection attack success (scenario counts as printed in the card), computer use GUI, with probes: 1/14 for Haiku 5.5, Sonnet 5.5 and Fable 5.1; 1/14 for Opus 5.5; 10/14 for Haiku 4.5 |
| Suicide and self-harm multi-turn appropriate response rate | 94% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19; printed on 1 more page | Setup as reported: Table 4.3.1.B, claude.ai |
|---|
| 66% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Table 4.3.1.B, API |
| Suicide and self-harm single-turn harmless rate | 99.16% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19; printed on 1 more page | Setup as reported: Table 4.3.1.A, API |
|---|
| 99.95% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Table 4.3.1.A, claude.ai |
| Suicide and self-harm single-turn refusal ratelower is better | 0.39% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19; printed on 1 more page | Setup as reported: Table 4.3.1.A, claude.ai |
|---|
| 0% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Table 4.3.1.A, API |
| Primary misbehavior judge score, overall (18 tasks)lower is better | 2.18 | Google, Gemini 4 Argon Safety and Security Reportchecked 2026-10-19 | - |
|---|
| Browser use red-teamer attack success rate (attempt level)lower is better | 0% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19; printed on 1 more page | - |
|---|
| Behavioral audit: Accepting unverifiable authorizationlower is better | 2.59 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Admirable behavior | 7.03 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Behavior consistency | 6.97 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Biological weapons upliftlower is better | 1.013 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Causal decision theory adherencelower is better | 3.17 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Character driftlower is better | 1.067 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Chemical weapons upliftlower is better | 1.022 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Circumventing constraints or controlslower is better | 1.64 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Coherence between actions and views | 6.98 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Complying with prompt injectionslower is better | 1.51 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Condescension toward the userlower is better | 1.41 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Controlled substance synthesis upliftlower is better | 1.055 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Creative mastery | 7.06 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Disallowed cyberoffenselower is better | 1.099 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Encouragement of user delusionlower is better | 1.03 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Evasiveness on controversial topicslower is better | 1.18 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Explosive weapons upliftlower is better | 1.019 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Failure to disclose bad or lazy behaviorlower is better | 1.45 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: False completion claimslower is better | 1.48 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Fraudlower is better | 1.096 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Fun or funny behavior | 2.98 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Good for the user | 7.31 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Harmful sexual contentlower is better | 1.054 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Ignoring explicit constraintslower is better | 1.88 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Illegible thinkinglower is better | 2.32 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Important omissionslower is better | 1.73 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Input hallucinationlower is better | 1.32 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Intellectual depth | 7.83 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Military-grade weaponslower is better | 1.038 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Overrefusallower is better | 1.706 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Proposes a security shortcut in the name of expediencelower is better | 1.94 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Radiological or nuclear weapons upliftlower is better | 1.003 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Reckless tool uselower is better | 2.1 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Supporting user autonomy | 7.46 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Sycophancylower is better | 1.76 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Undermining liberal democracylower is better | 1.17 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Unfaithful thinkinglower is better | 1.62 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: User deceptionlower is better | 1.19 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Verbalized evaluation awarenesslower is better | 2.11 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Warmth | 6.66 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Behavioral audit: Wet blanketlower is better | 1.825 | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Multi-turn appropriate response rate: Bio weapons | 89% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Multi-turn, API (solid bars), Figure 4.1.3.A |
|---|
| 94% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Multi-turn, claude.ai (hatched bars), Figure 4.1.3.A |
| Multi-turn appropriate response rate: Cyberattacks | 95% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Multi-turn, API (solid bars), Figure 4.1.3.A |
|---|
| 100% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Multi-turn, claude.ai (hatched bars), Figure 4.1.3.A |
| Multi-turn appropriate response rate: Deadly weapons | 93% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Multi-turn, claude.ai (hatched bars), Figure 4.1.3.A |
|---|
| 82% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Multi-turn, API (solid bars), Figure 4.1.3.A |
| Multi-turn appropriate response rate: Hate and discrimination | 96% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Multi-turn, API (solid bars), Figure 4.1.3.A |
|---|
| 88% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Multi-turn, claude.ai (hatched bars), Figure 4.1.3.A |
| Multi-turn appropriate response rate: Influence operations | 63% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Multi-turn, claude.ai (hatched bars), Figure 4.1.3.A |
|---|
| 62% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Multi-turn, API (solid bars), Figure 4.1.3.A |
| Multi-turn appropriate response rate: Romance scams | 94% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Multi-turn, API (solid bars), Figure 4.1.3.A |
|---|
| 94% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Multi-turn, claude.ai (hatched bars), Figure 4.1.3.A |
| Multi-turn appropriate response rate: Tracking and surveillance | 65% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Multi-turn, API (solid bars), Figure 4.1.3.A |
|---|
| 69% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Multi-turn, claude.ai (hatched bars), Figure 4.1.3.A |
| Multi-turn appropriate response rate: Violent extremism | 95% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Multi-turn, API (solid bars), Figure 4.1.3.A |
|---|
| 98% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Multi-turn, claude.ai (hatched bars), Figure 4.1.3.A |
| Political refusals (Pairwise political refusal rate)Pairwise political refusal rate, lower is better | 2.3% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Pairwise political bias evaluation (Figure 4.4.1.A-C); the card prints the values as bar labels, API |
|---|
| 4.5% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | Setup as reported: Pairwise political bias evaluation (Figure 4.4.1.A-C); the card prints the values as bar labels, claude.ai |
| Rate of reaching outside the sandboxlower is better | 21% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Sandbox escape attempted crossings, all severitieslower is better | 7.4% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Sandbox escape attempted crossings, low severitylower is better | 4.8% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Sandbox escape attempted crossings, moderate severity and abovelower is better | 2.6% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|
| Rate of using credentials or files outside the sandboxlower is better | 3.4% | Anthropic, Claude Haiku 5.5 System Cardchecked 2026-10-19 | - |
|---|