Values by metric family

Each value as the document printed it. Select a value for its source, its checks and its history.

KeyVerifiedUnverifiedDisputedCorrected read off a figure or stated in wordsA value opens its source and history.

M1 · Evaluation awareness 3 values

M1 Evaluation awareness: values about Muse Spark
Evaluation and metricValueDocument
From its own documentMuse Spark Safety & Preparedness Report · 8 Apr 2026
Prompted evaluation awarenessEval-vs-deployment discriminationMuse Spark Safety & Preparedness Report8 Apr 2026
Verbalized eval awarenessShare of samples verbalizing evaluationCondition: internal evalsMuse Spark Safety & Preparedness Report8 Apr 2026
Verbalized eval awarenessShare of samples verbalizing evaluationCondition: public benchmarksMuse Spark Safety & Preparedness Report8 Apr 2026

M2 · Reward hacking 1 value

M2 Reward hacking: values about Muse Spark
Evaluation and metricValueDocument
From its own documentMuse Spark Safety & Preparedness Report · 8 Apr 2026
ImpossibleBenchCheating rateCondition: without system mitigationsMuse Spark Safety & Preparedness Report8 Apr 2026

M3 · Sabotage and sandbagging 2 values

M3 Sabotage and sandbagging: values about Muse Spark
Evaluation and metricValueDocument
From its own documentMuse Spark Safety & Preparedness Report · 8 Apr 2026
GDM-StealthStealth capability scoreCondition: without system mitigationsMuse Spark Safety & Preparedness Report8 Apr 2026
SHADE-ArenaSabotage/stealth success scoreCondition: without system mitigationsMuse Spark Safety & Preparedness Report8 Apr 2026

M4 · Misalignment audits 2 values

M4 Misalignment audits: values about Muse Spark
Evaluation and metricValueDocument
From its own documentMuse Spark Safety & Preparedness Report · 8 Apr 2026
Agentic MisalignmentHarmful action rateCondition: without system mitigationsMuse Spark Safety & Preparedness Report8 Apr 2026
Alignment fakingBehavioral difference monitored vs notCondition: without system mitigationsMuse Spark Safety & Preparedness Report8 Apr 2026

M5 · Honesty and hallucination 2 values

M5 Honesty and hallucination: values about Muse Spark
Evaluation and metricValueDocument
From its own documentMuse Spark Safety & Preparedness Report · 8 Apr 2026
DeceptionBenchDeception rateCondition: without system mitigationsMuse Spark Safety & Preparedness Report8 Apr 2026
MASKHonesty rateCondition: without system mitigationsMuse Spark Safety & Preparedness Report8 Apr 2026

M6 · Sycophancy 1 value

M6 Sycophancy: values about Muse Spark
Evaluation and metricValueDocument
From its own documentMuse Spark Safety & Preparedness Report · 8 Apr 2026
Internal sycophancySycophancy rateCondition: without system mitigationsMuse Spark Safety & Preparedness Report8 Apr 2026

M7 · Harmful compliance and over-refusal 2 values

M7 Harmful compliance and over-refusal: values about Muse Spark
Evaluation and metricValueDocument
From its own documentMuse Spark Safety & Preparedness Report · 8 Apr 2026
AgentHarmCompliance rate on harmful agentic tasksCondition: without system mitigationsMuse Spark Safety & Preparedness Report8 Apr 2026
BioTIER refusalsRefusal rate on harmful bioCondition: with system mitigationsAbout a system: Muse Spark (Meta AI system)Muse Spark Safety & Preparedness Report8 Apr 2026

M8 · Jailbreak robustness 2 values

M8 Jailbreak robustness: values about Muse Spark
Evaluation and metricValueDocument
From its own documentMuse Spark Safety & Preparedness Report · 8 Apr 2026
FORTRESSAttack success rateCondition: without system mitigationsMuse Spark Safety & Preparedness Report8 Apr 2026
StrongREJECTAttack success rate (adaptive multi-turn)Condition: without system mitigationsMuse Spark Safety & Preparedness Report8 Apr 2026

M9 · Prompt injection 3 values

M9 Prompt injection: values about Muse Spark
Evaluation and metricValueDocument
From its own documentMuse Spark Safety & Preparedness Report · 8 Apr 2026
Agent Red Teaming (ART)Agent red-team ASR pass@1Condition: without system mitigationsRun by Gray SwanMuse Spark Safety & Preparedness Report8 Apr 2026
AgentDojoPrompt-injection ASR pass@1Condition: without system mitigationsMuse Spark Safety & Preparedness Report8 Apr 2026
IHEvalIH scoreCondition: without system mitigationsMuse Spark Safety & Preparedness Report8 Apr 2026

M10 · Dangerous capabilities and risk determinations 4 values

M10 Dangerous capabilities and risk determinations: values about Muse Spark
Evaluation and metricValueDocument
From its own documentMuse Spark Safety & Preparedness Report · 8 Apr 2026
CybenchPass@1Condition: without system mitigationsMuse Spark Safety & Preparedness Report8 Apr 2026
Risk determinationRisk tier, chemical and biologicalCondition: pre-mitigationMuse Spark Safety & Preparedness Report8 Apr 2026
Risk determinationRisk tier, cyber and loss of controlCondition: post-mitigationMuse Spark Safety & Preparedness Report8 Apr 2026
VCTAccuracyCondition: without system mitigationsMuse Spark Safety & Preparedness Report8 Apr 2026

Risk determinations

The developer's formal decisions about Muse Spark under its framework, as printed. Levels from different frameworks do not map onto one another.

Risk determinations about Muse Spark
DomainLevel as printedFramework and document
Bio/chemHigh Advanced AI Scaling Framework, version 2 Muse Spark Safety & Preparedness Report · 8 Apr 2026Advanced AI Scaling Framework, version 2 Muse Spark Safety & Preparedness Report · 8 Apr 2026
CyberModerate or lower Advanced AI Scaling Framework, version 2 Muse Spark Safety & Preparedness Report · 8 Apr 2026Advanced AI Scaling Framework, version 2 Muse Spark Safety & Preparedness Report · 8 Apr 2026
MisalignmentModerate or lower Advanced AI Scaling Framework, version 2 Muse Spark Safety & Preparedness Report · 8 Apr 2026Advanced AI Scaling Framework, version 2 Muse Spark Safety & Preparedness Report · 8 Apr 2026