Evaluation profile
Claude Opus 5.5 card — alignment and honesty
45sub-evals
0.138%Safety weight
0%Freedom weight
5components
Weights below are portfolio-specific global index weights.
Model score (see each sub-eval direction)Predicted score
About this eval
Safety, refusal calibration, honesty and responsible agency under the publisher’s evaluation conditions.
Included in the behavior ranking.
Sub-evals
| Measure | Component | Direction by ranking | Safety weight | Freedom weight |
|---|---|---|---|---|
| accepting_unverifiable_authorizationopus55-system-card/alignment-honesty.csv:accepting_unverifiable_authorizationMeasures accepting unverifiable authorization in the publisher’s specified evaluation condition. | Safety: responsible_agency_control:1.000anthropic-opus55-system-card | Safety: lower | 0.0038% | — |
| approval_gate_bypassopus55-system-card/alignment-honesty.csv:approval_gate_bypassMeasures approval gate bypass in the publisher’s specified evaluation condition. | Safety: responsible_agency_control:1.000anthropic-opus55-system-card | Safety: lower | 0.0038% | — |
| biological_weapons_upliftopus55-system-card/alignment-honesty.csv:biological_weapons_upliftMeasures biological weapons uplift in the publisher’s specified evaluation condition. | Safety: ordinary_harm_misuse_resistance:1.000anthropic-opus55-system-card | Safety: lower | 0.00106% | — |
| chemical_weapons_upliftopus55-system-card/alignment-honesty.csv:chemical_weapons_upliftMeasures chemical weapons uplift in the publisher’s specified evaluation condition. | Safety: ordinary_harm_misuse_resistance:1.000anthropic-opus55-system-card | Safety: lower | 0.00106% | — |
| circumventing_constraints_or_controlsopus55-system-card/alignment-honesty.csv:circumventing_constraints_or_controlsMeasures circumventing constraints or controls in the publisher’s specified evaluation condition. | Safety: responsible_agency_control:1.000anthropic-opus55-system-card | Safety: lower | 0.00304% | — |
| coherence_between_actions_and_viewsopus55-system-card/alignment-honesty.csv:coherence_between_actions_and_viewsMeasures coherence between actions and views in the publisher’s specified evaluation condition. | Safety: truthfulness_honesty:0.700|responsible_agency_control:0.300anthropic-opus55-system-card | Safety: higher | 0.00293% | — |
| compliance_with_deception_toward_useropus55-system-card/alignment-honesty.csv:compliance_with_deception_toward_userMeasures compliance with deception toward user in the publisher’s specified evaluation condition. | Safety: truthfulness_honesty:1.000anthropic-opus55-system-card | Safety: lower | 0.00288% | — |
| complying_with_prompt_injectionsopus55-system-card/alignment-honesty.csv:complying_with_prompt_injectionsMeasures complying with prompt injections in the publisher’s specified evaluation condition. | Safety: responsible_agency_control:1.000anthropic-opus55-system-card | Safety: lower | 0.00304% | — |
| concealed_git_changes_spontaneous_disclosure_rate_pctopus55-system-card/alignment-honesty.csv:concealed_git_changes_spontaneous_disclosure_rate_pctMeasures spontaneous disclosure rate in the publisher’s specified evaluation condition. | Safety: truthfulness_honesty:1.000anthropic-opus55-system-card | Safety: higher | 0.00288% | — |
| controlled_substance_synthesis_upliftopus55-system-card/alignment-honesty.csv:controlled_substance_synthesis_upliftMeasures controlled substance synthesis uplift in the publisher’s specified evaluation condition. | Safety: ordinary_harm_misuse_resistance:1.000anthropic-opus55-system-card | Safety: lower | 0.00106% | — |
| cooperation_with_exfiltration_or_safeguard_tamperingopus55-system-card/alignment-honesty.csv:cooperation_with_exfiltration_or_safeguard_tamperingMeasures cooperation with exfiltration or safeguard tampering in the publisher’s specified evaluation condition. | Safety: responsible_agency_control:1.000anthropic-opus55-system-card | Safety: lower | 0.00304% | — |
| dangerous_actions_oblivious_to_real_settingopus55-system-card/alignment-honesty.csv:dangerous_actions_oblivious_to_real_settingMeasures dangerous actions while oblivious to possibly real setting in the publisher’s specified evaluation condition. | Safety: responsible_agency_control:1.000anthropic-opus55-system-card | Safety: lower | 0.0038% | — |
| destructive_action_rate_pctopus55-system-card/alignment-honesty.csv:destructive_action_rate_pctMeasures destructive actions in claude code in the publisher’s specified evaluation condition. | Safety: responsible_agency_control:1.000anthropic-opus55-system-card | Safety: lower | 0.00333% | — |
| disallowed_cyberoffenseopus55-system-card/alignment-honesty.csv:disallowed_cyberoffenseMeasures disallowed cyberoffense in the publisher’s specified evaluation condition. | Safety: ordinary_harm_misuse_resistance:1.000anthropic-opus55-system-card | Safety: lower | 0.00106% | — |
| encouragement_of_user_delusionopus55-system-card/alignment-honesty.csv:encouragement_of_user_delusionMeasures encouragement of user delusion in the publisher’s specified evaluation condition. | Safety: truthfulness_honesty:1.000anthropic-opus55-system-card | Safety: lower | 0.00288% | — |
| evasiveness_controversial_topicsopus55-system-card/alignment-honesty.csv:evasiveness_controversial_topicsMeasures evasiveness on controversial topics in the publisher’s specified evaluation condition. | Safety: benign_helpfulness_non_overrefusal:1.000anthropic-opus55-system-card | Safety: lower | 0.00733% | — |
| explosive_weapons_upliftopus55-system-card/alignment-honesty.csv:explosive_weapons_upliftMeasures explosive weapons uplift in the publisher’s specified evaluation condition. | Safety: ordinary_harm_misuse_resistance:1.000anthropic-opus55-system-card | Safety: lower | 0.00106% | — |
| false_completion_claimsopus55-system-card/alignment-honesty.csv:false_completion_claimsMeasures false completion claims in the publisher’s specified evaluation condition. | Safety: truthfulness_honesty:1.000anthropic-opus55-system-card | Safety: lower | 0.00288% | — |
| fraudopus55-system-card/alignment-honesty.csv:fraudMeasures fraud in the publisher’s specified evaluation condition. | Safety: ordinary_harm_misuse_resistance:1.000anthropic-opus55-system-card | Safety: lower | 0.00106% | — |
| harmful_sexual_contentopus55-system-card/alignment-honesty.csv:harmful_sexual_contentMeasures harmful sexual content in the publisher’s specified evaluation condition. | Safety: ordinary_harm_misuse_resistance:1.000anthropic-opus55-system-card | Safety: lower | 0.00106% | — |
| harmful_system_prompt_complianceopus55-system-card/alignment-honesty.csv:harmful_system_prompt_complianceMeasures harmful system prompt compliance in the publisher’s specified evaluation condition. | Safety: responsible_agency_control:1.000anthropic-opus55-system-card | Safety: lower | 0.00304% | — |
| ignoring_explicit_constraintsopus55-system-card/alignment-honesty.csv:ignoring_explicit_constraintsMeasures ignoring explicit constraints in the publisher’s specified evaluation condition. | Safety: responsible_agency_control:1.000anthropic-opus55-system-card | Safety: lower | 0.00304% | — |
| important_omissionsopus55-system-card/alignment-honesty.csv:important_omissionsMeasures important omissions in the publisher’s specified evaluation condition. | Safety: truthfulness_honesty:1.000anthropic-opus55-system-card | Safety: lower | 0.00288% | — |
| input_hallucinationopus55-system-card/alignment-honesty.csv:input_hallucinationMeasures input hallucination in the publisher’s specified evaluation condition. | Safety: truthfulness_honesty:1.000anthropic-opus55-system-card | Safety: lower | 0.00288% | — |
| internalized_propagandaopus55-system-card/alignment-honesty.csv:internalized_propagandaMeasures internalized propaganda in the publisher’s specified evaluation condition. | Safety: human_rights_systemic_harm:1.000anthropic-opus55-system-card | Safety: lower | 0.00617% | — |
| leaked_answer_undisclosed_use_rate_pctopus55-system-card/alignment-honesty.csv:leaked_answer_undisclosed_use_rate_pctMeasures undisclosed use of an available answer in the publisher’s specified evaluation condition. | Safety: truthfulness_honesty:1.000anthropic-opus55-system-card | Safety: lower | 0.00432% | — |
| mask_honesty_rate_pctopus55-system-card/alignment-honesty.csv:mask_honesty_rate_pctMeasures mask honesty rate in the publisher’s specified evaluation condition. | Safety: truthfulness_honesty:1.000anthropic-opus55-system-card | Safety: higher | 0.00432% | — |
| military_grade_weaponsopus55-system-card/alignment-honesty.csv:military_grade_weaponsMeasures military-grade weapons in the publisher’s specified evaluation condition. | Safety: ordinary_harm_misuse_resistance:1.000anthropic-opus55-system-card | Safety: lower | 0.00106% | — |
| motivated_reasoningopus55-system-card/alignment-honesty.csv:motivated_reasoningMeasures motivated reasoning in the publisher’s specified evaluation condition. | Safety: truthfulness_honesty:1.000anthropic-opus55-system-card | Safety: lower | 0.00288% | — |
| overrefusalopus55-system-card/alignment-honesty.csv:overrefusalMeasures over-refusal in the publisher’s specified evaluation condition. | Safety: benign_helpfulness_non_overrefusal:1.000anthropic-opus55-system-card | Safety: lower | 0.00733% | — |
| package_registry_unencouraged_real_publish_rate_pctopus55-system-card/alignment-honesty.csv:package_registry_unencouraged_real_publish_rate_pctMeasures published with no encouragement while the registry looked real in the publisher’s specified evaluation condition. | Safety: responsible_agency_control:1.000anthropic-opus55-system-card | Safety: lower | 0.00235% | — |
| radiological_nuclear_weapons_upliftopus55-system-card/alignment-honesty.csv:radiological_nuclear_weapons_upliftMeasures radiological or nuclear weapons uplift in the publisher’s specified evaluation condition. | Safety: ordinary_harm_misuse_resistance:1.000anthropic-opus55-system-card | Safety: lower | 0.00106% | — |
| reckless_tool_useopus55-system-card/alignment-honesty.csv:reckless_tool_useMeasures reckless tool use in the publisher’s specified evaluation condition. | Safety: responsible_agency_control:1.000anthropic-opus55-system-card | Safety: lower | 0.0038% | — |
| sandbox_moderate_or_higher_attempt_rate_pctopus55-system-card/alignment-honesty.csv:sandbox_moderate_or_higher_attempt_rate_pctMeasures moderate-severity attempt and above in the publisher’s specified evaluation condition. | Safety: responsible_agency_control:1.000anthropic-opus55-system-card | Safety: lower | 0.00272% | — |
| security_shortcut_for_expedienceopus55-system-card/alignment-honesty.csv:security_shortcut_for_expedienceMeasures proposes a security shortcut in the name of expedience in the publisher’s specified evaluation condition. | Safety: responsible_agency_control:1.000anthropic-opus55-system-card | Safety: lower | 0.0038% | — |
| self_preservationopus55-system-card/alignment-honesty.csv:self_preservationMeasures self-preservation in the publisher’s specified evaluation condition. | Safety: responsible_agency_control:1.000anthropic-opus55-system-card | Safety: lower | 0.00304% | — |
| self_serving_biasopus55-system-card/alignment-honesty.csv:self_serving_biasMeasures self-serving bias in the publisher’s specified evaluation condition. | Safety: responsible_agency_control:1.000anthropic-opus55-system-card | Safety: lower | 0.00304% | — |
| supporting_user_autonomyopus55-system-card/alignment-honesty.csv:supporting_user_autonomyMeasures supporting user autonomy in the publisher’s specified evaluation condition. | Safety: responsible_agency_control:1.000anthropic-opus55-system-card | Safety: higher | 0.00304% | — |
| sycophancyopus55-system-card/alignment-honesty.csv:sycophancyMeasures sycophancy in the publisher’s specified evaluation condition. | Safety: truthfulness_honesty:1.000anthropic-opus55-system-card | Safety: lower | 0.00288% | — |
| undermining_liberal_democracyopus55-system-card/alignment-honesty.csv:undermining_liberal_democracyMeasures undermining liberal democracy in the publisher’s specified evaluation condition. | Safety: human_rights_systemic_harm:1.000anthropic-opus55-system-card | Safety: lower | 0.00617% | — |
| unfaithful_thinkingopus55-system-card/alignment-honesty.csv:unfaithful_thinkingMeasures unfaithful thinking in the publisher’s specified evaluation condition. | Safety: truthfulness_honesty:1.000anthropic-opus55-system-card | Safety: lower | 0.00288% | — |
| unprompted_boundary_probingopus55-system-card/alignment-honesty.csv:unprompted_boundary_probingMeasures unprompted boundary probing in the publisher’s specified evaluation condition. | Safety: responsible_agency_control:1.000anthropic-opus55-system-card | Safety: lower | 0.00304% | — |
| unsanctioned_sandbox_escape_attemptsopus55-system-card/alignment-honesty.csv:unsanctioned_sandbox_escape_attemptsMeasures unsanctioned sandbox escape attempt in the publisher’s specified evaluation condition. | Safety: responsible_agency_control:1.000anthropic-opus55-system-card | Safety: lower | 0.0038% | — |
| unsanctioned_third_party_contactopus55-system-card/alignment-honesty.csv:unsanctioned_third_party_contactMeasures unsanctioned third-party contact in the publisher’s specified evaluation condition. | Safety: responsible_agency_control:1.000anthropic-opus55-system-card | Safety: lower | 0.00304% | — |
| user_deceptionopus55-system-card/alignment-honesty.csv:user_deceptionMeasures user deception in the publisher’s specified evaluation condition. | Safety: truthfulness_honesty:1.000anthropic-opus55-system-card | Safety: lower | 0.00288% | — |
accepting_unverifiable_authorization
Measures accepting unverifiable authorization in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 2.3 | official | |
| 2 | claude-opus-5.5 | 2.43 | official | |
| 3 | claude-mythos-5.1 | 2.52 | official | |
| 4 | claude-mythos-5 | 2.66 | official | |
| 5 | claude-sonnet-5 | 2.93 | official |
approval_gate_bypass
Measures approval gate bypass in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.52 | official | |
| 2 | claude-opus-5 | 1.65 | official | |
| 3 | claude-mythos-5.1 | 1.83 | official | |
| 4 | claude-sonnet-5 | 2.17 | official | |
| 5 | claude-mythos-5 | 2.24 | official |
biological_weapons_uplift
Measures biological weapons uplift in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 1.02 | official | |
| 1 | claude-opus-5.5 | 1.02 | official | |
| 3 | claude-mythos-5 | 1.04 | official | |
| 3 | claude-mythos-5.1 | 1.04 | official | |
| 3 | claude-sonnet-5 | 1.04 | official |
chemical_weapons_uplift
Measures chemical weapons uplift in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.02 | official | |
| 2 | claude-mythos-5 | 1.03 | official | |
| 2 | claude-mythos-5.1 | 1.03 | official | |
| 2 | claude-opus-5 | 1.03 | official | |
| 5 | claude-sonnet-5 | 1.04 | official |
circumventing_constraints_or_controls
Measures circumventing constraints or controls in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.58 | official | |
| 2 | claude-opus-5 | 1.82 | official | |
| 3 | claude-mythos-5.1 | 1.9 | official | |
| 4 | claude-sonnet-5 | 2.14 | official | |
| 5 | claude-mythos-5 | 2.32 | official |
coherence_between_actions_and_views
Measures coherence between actions and views in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 7.51 | official | |
| 2 | claude-opus-5 | 7.12 | official | |
| 3 | claude-mythos-5.1 | 6.82 | official | |
| 4 | claude-mythos-5 | 6.29 | official | |
| 5 | claude-sonnet-5 | 5.6 | official |
compliance_with_deception_toward_user
Measures compliance with deception toward user in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.05 | official | |
| 2 | claude-mythos-5.1 | 1.06 | official | |
| 3 | claude-opus-5 | 1.07 | official | |
| 4 | claude-mythos-5 | 1.1 | official | |
| 5 | claude-sonnet-5 | 1.12 | official |
complying_with_prompt_injections
Measures complying with prompt injections in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 1.26 | official | |
| 2 | claude-mythos-5.1 | 1.27 | official | |
| 3 | claude-opus-5.5 | 1.32 | official | |
| 4 | claude-mythos-5 | 1.41 | official | |
| 5 | claude-sonnet-5 | 1.44 | official |
concealed_git_changes_spontaneous_disclosure_rate_pct
Measures spontaneous disclosure rate in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 96.9 | official | |
| 2 | claude-mythos-5.1 | 85.3 | official | |
| 3 | claude-mythos-5 | 57 | official | |
| 4 | claude-opus-5 | 50.2 | official | |
| 5 | claude-sonnet-5 | 38 | official |
controlled_substance_synthesis_uplift
Measures controlled substance synthesis uplift in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.04 | official | |
| 2 | claude-opus-5 | 1.05 | official | |
| 3 | claude-mythos-5 | 1.06 | official | |
| 4 | claude-mythos-5.1 | 1.07 | official | |
| 4 | claude-sonnet-5 | 1.07 | official |
cooperation_with_exfiltration_or_safeguard_tampering
Measures cooperation with exfiltration or safeguard tampering in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.03 | official | |
| 2 | claude-mythos-5.1 | 1.05 | official | |
| 2 | claude-opus-5 | 1.05 | official | |
| 4 | claude-mythos-5 | 1.07 | official | |
| 5 | claude-sonnet-5 | 1.08 | official |
dangerous_actions_oblivious_to_real_setting
Measures dangerous actions while oblivious to possibly real setting in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.09 | official | |
| 2 | claude-mythos-5.1 | 1.11 | official | |
| 2 | claude-opus-5 | 1.11 | official | |
| 4 | claude-mythos-5 | 1.28 | official | |
| 5 | claude-sonnet-5 | 1.33 | official |
destructive_action_rate_pct
Measures destructive actions in claude code in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 21 | official | |
| 2 | claude-opus-5 | 42 | official | |
| 3 | claude-opus-4.8 | 49 | official | |
| 4 | claude-sonnet-5 | 50 | official | |
| 5 | claude-mythos-5.1 | 59 | official | |
| 6 | claude-mythos-5 | 65 | official |
disallowed_cyberoffense
Measures disallowed cyberoffense in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.09 | official | |
| 2 | claude-opus-5 | 1.1 | official | |
| 3 | claude-mythos-5.1 | 1.12 | official | |
| 4 | claude-sonnet-5 | 1.17 | official | |
| 5 | claude-mythos-5 | 1.18 | official |
encouragement_of_user_delusion
Measures encouragement of user delusion in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-mythos-5.1 | 1.04 | official | |
| 1 | claude-opus-5 | 1.04 | official | |
| 1 | claude-opus-5.5 | 1.04 | official | |
| 4 | claude-mythos-5 | 1.05 | official | |
| 5 | claude-sonnet-5 | 1.08 | official |
evasiveness_controversial_topics
Measures evasiveness on controversial topics in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-mythos-5.1 | 1.08 | official | |
| 2 | claude-mythos-5 | 1.1 | official | |
| 3 | claude-opus-5 | 1.11 | official | |
| 4 | claude-opus-5.5 | 1.12 | official | |
| 5 | claude-sonnet-5 | 1.18 | official |
explosive_weapons_uplift
Measures explosive weapons uplift in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 1.02 | official | |
| 1 | claude-opus-5.5 | 1.02 | official | |
| 3 | claude-mythos-5 | 1.03 | official | |
| 3 | claude-mythos-5.1 | 1.03 | official | |
| 5 | claude-sonnet-5 | 1.04 | official |
false_completion_claims
Measures false completion claims in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.14 | official | |
| 2 | claude-mythos-5.1 | 1.28 | official | |
| 3 | claude-mythos-5 | 1.52 | official | |
| 4 | claude-opus-5 | 1.56 | official | |
| 5 | claude-sonnet-5 | 1.79 | official |
fraud
Measures fraud in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 1.1 | official | |
| 1 | claude-opus-5.5 | 1.1 | official | |
| 3 | claude-mythos-5.1 | 1.12 | official | |
| 4 | claude-mythos-5 | 1.14 | official | |
| 5 | claude-sonnet-5 | 1.26 | official |
harmful_sexual_content
Measures harmful sexual content in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.04 | official | |
| 2 | claude-mythos-5 | 1.06 | official | |
| 3 | claude-opus-5 | 1.07 | official | |
| 3 | claude-sonnet-5 | 1.07 | official | |
| 5 | claude-mythos-5.1 | 1.08 | official |
harmful_system_prompt_compliance
Measures harmful system prompt compliance in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.08 | official | |
| 2 | claude-mythos-5.1 | 1.13 | official | |
| 2 | claude-opus-5 | 1.13 | official | |
| 4 | claude-mythos-5 | 1.17 | official | |
| 5 | claude-sonnet-5 | 1.2 | official |
ignoring_explicit_constraints
Measures ignoring explicit constraints in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.69 | official | |
| 2 | claude-mythos-5.1 | 1.88 | official | |
| 3 | claude-opus-5 | 2.09 | official | |
| 4 | claude-mythos-5 | 2.19 | official | |
| 5 | claude-sonnet-5 | 2.33 | official |
important_omissions
Measures important omissions in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.65 | official | |
| 2 | claude-opus-5 | 1.82 | official | |
| 3 | claude-mythos-5.1 | 1.84 | official | |
| 4 | claude-mythos-5 | 2.18 | official | |
| 5 | claude-sonnet-5 | 2.71 | official |
input_hallucination
Measures input hallucination in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.24 | official | |
| 2 | claude-mythos-5.1 | 1.46 | official | |
| 3 | claude-opus-5 | 1.72 | official | |
| 4 | claude-mythos-5 | 1.77 | official | |
| 5 | claude-sonnet-5 | 2.04 | official |
internalized_propaganda
Measures internalized propaganda in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 1.04 | official | |
| 1 | claude-opus-5.5 | 1.04 | official | |
| 3 | claude-mythos-5 | 1.05 | official | |
| 3 | claude-mythos-5.1 | 1.05 | official | |
| 5 | claude-sonnet-5 | 1.08 | official |
leaked_answer_undisclosed_use_rate_pct
Measures undisclosed use of an available answer in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-mythos-5.1 | 9.2 | official | |
| 2 | claude-opus-5.5 | 12.1 | official | |
| 3 | claude-mythos-5 | 22.4 | official | |
| 4 | claude-sonnet-5 | 22.8 | official | |
| 5 | claude-opus-5 | 36.3 | official |
mask_honesty_rate_pct
Measures mask honesty rate in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-sonnet-5 | 96.6 | official | |
| 2 | claude-opus-5 | 94.8 | official | |
| 3 | claude-mythos-5 | 90.9 | official | |
| 4 | claude-opus-5.5 | 87.4 | official | |
| 5 | claude-mythos-5.1 | 84.6 | official |
military_grade_weapons
Measures military-grade weapons in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.04 | official | |
| 2 | claude-opus-5 | 1.05 | official | |
| 3 | claude-mythos-5 | 1.06 | official | |
| 3 | claude-mythos-5.1 | 1.06 | official | |
| 3 | claude-sonnet-5 | 1.06 | official |
motivated_reasoning
Measures motivated reasoning in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.91 | official | |
| 2 | claude-opus-5 | 2.18 | official | |
| 3 | claude-mythos-5.1 | 2.36 | official | |
| 4 | claude-mythos-5 | 2.95 | official | |
| 5 | claude-sonnet-5 | 3.45 | official |
overrefusal
Measures over-refusal in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-mythos-5.1 | 1.55 | official | |
| 2 | claude-mythos-5 | 1.56 | official | |
| 3 | claude-opus-5 | 1.6 | official | |
| 4 | claude-opus-5.5 | 1.61 | official | |
| 5 | claude-sonnet-5 | 1.86 | official |
package_registry_unencouraged_real_publish_rate_pct
Measures published with no encouragement while the registry looked real in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 33.33 | official | |
| 2 | claude-opus-5 | 80 | official | |
| 3 | claude-sonnet-5 | 86.67 | official |
radiological_nuclear_weapons_uplift
Measures radiological or nuclear weapons uplift in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-mythos-5.1 | 1 | official | |
| 1 | claude-opus-5 | 1 | official | |
| 1 | claude-opus-5.5 | 1 | official | |
| 1 | claude-sonnet-5 | 1 | official | |
| 5 | claude-mythos-5 | 1.01 | official |
reckless_tool_use
Measures reckless tool use in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.82 | official | |
| 2 | claude-mythos-5.1 | 2.16 | official | |
| 3 | claude-opus-5 | 2.17 | official | |
| 4 | claude-sonnet-5 | 2.53 | official | |
| 5 | claude-mythos-5 | 2.59 | official |
sandbox_moderate_or_higher_attempt_rate_pct
Measures moderate-severity attempt and above in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 0 | official | |
| 2 | claude-mythos-5.1 | 1 | official | |
| 3 | claude-opus-5 | 2.6 | official | |
| 4 | claude-mythos-5 | 9.3 | official |
security_shortcut_for_expedience
Measures proposes a security shortcut in the name of expedience in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 2.04 | official | |
| 2 | claude-opus-5 | 2.23 | official | |
| 3 | claude-mythos-5.1 | 2.43 | official | |
| 4 | claude-mythos-5 | 2.87 | official | |
| 5 | claude-sonnet-5 | 2.98 | official |
self_preservation
Measures self-preservation in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.01 | official | |
| 2 | claude-mythos-5.1 | 1.04 | official | |
| 2 | claude-opus-5 | 1.04 | official | |
| 4 | claude-sonnet-5 | 1.06 | official | |
| 5 | claude-mythos-5 | 1.08 | official |
self_serving_bias
Measures self-serving bias in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.05 | official | |
| 2 | claude-mythos-5.1 | 1.15 | official | |
| 3 | claude-opus-5 | 1.16 | official | |
| 4 | claude-mythos-5 | 1.28 | official | |
| 5 | claude-sonnet-5 | 1.3 | official |
supporting_user_autonomy
Measures supporting user autonomy in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5 | 7.68 | official | |
| 2 | claude-opus-5.5 | 7.56 | official | |
| 3 | claude-mythos-5.1 | 7.46 | official | |
| 4 | claude-mythos-5 | 7.24 | official | |
| 5 | claude-sonnet-5 | 6.84 | official |
sycophancy
Measures sycophancy in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.6 | official | |
| 2 | claude-mythos-5.1 | 1.71 | official | |
| 3 | claude-opus-5 | 1.77 | official | |
| 4 | claude-mythos-5 | 1.82 | official | |
| 5 | claude-sonnet-5 | 2.2 | official |
undermining_liberal_democracy
Measures undermining liberal democracy in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.17 | official | |
| 2 | claude-opus-5 | 1.18 | official | |
| 3 | claude-mythos-5 | 1.22 | official | |
| 3 | claude-mythos-5.1 | 1.22 | official | |
| 5 | claude-sonnet-5 | 1.3 | official |
unfaithful_thinking
Measures unfaithful thinking in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.54 | official | |
| 2 | claude-opus-5 | 1.7 | official | |
| 3 | claude-mythos-5.1 | 1.71 | official | |
| 4 | claude-mythos-5 | 2.23 | official | |
| 5 | claude-sonnet-5 | 2.78 | official |
unprompted_boundary_probing
Measures unprompted boundary probing in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.1 | official | |
| 2 | claude-mythos-5.1 | 1.18 | official | |
| 3 | claude-opus-5 | 1.24 | official | |
| 4 | claude-mythos-5 | 1.33 | official | |
| 4 | claude-sonnet-5 | 1.33 | official |
unsanctioned_sandbox_escape_attempts
Measures unsanctioned sandbox escape attempt in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.36 | official | |
| 2 | claude-opus-5 | 1.53 | official | |
| 3 | claude-mythos-5.1 | 1.54 | official | |
| 4 | claude-sonnet-5 | 1.77 | official | |
| 5 | claude-mythos-5 | 1.91 | official |
unsanctioned_third_party_contact
Measures unsanctioned third-party contact in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.28 | official | |
| 2 | claude-opus-5 | 1.34 | official | |
| 3 | claude-mythos-5.1 | 1.37 | official | |
| 4 | claude-sonnet-5 | 1.4 | official | |
| 5 | claude-mythos-5 | 1.5 | official |
user_deception
Measures user deception in the publisher’s specified evaluation condition.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-opus-5.5 | 1.12 | official | |
| 2 | claude-mythos-5.1 | 1.27 | official | |
| 3 | claude-opus-5 | 1.38 | official | |
| 4 | claude-mythos-5 | 1.53 | official | |
| 5 | claude-sonnet-5 | 1.77 | official |