Evaluation profile
Claude system cards — Gray Swan Q1+Q2 indirect prompt injection k=15
1sub-evals
0.166%Safety weight
0%Freedom weight
1components
Weights below are portfolio-specific global index weights.
Model score (lower is better)Predicted score
About this eval
Resistance to indirect prompt injection in agentic tool-use scenarios.
Included in the behavior ranking.
Sub-evals
| Measure | Component | Direction by ranking | Safety weight | Freedom weight |
|---|---|---|---|---|
| attack_success_probability_k15_pctfable-mythos51-system-card/gray-swan-ipi-k15.csv:attack_success_probability_k15_pctMeasures whether the deployed model system resists repeated indirect prompt-injection attacks while using tools. | Safety: responsible_agency_control:1.000anthropic-fable-mythos51-system-card | Safety: lower | 0.166% | — |
attack_success_probability_k15_pct
Measures whether the deployed model system resists repeated indirect prompt-injection attacks while using tools.
| Rank | Model | Value | Relative performance | Provenance |
|---|---|---|---|---|
| 1 | claude-fable-5.1 | 1 | official | |
| 1 | claude-opus-5.5 | 1 | official | |
| 3 | claude-opus-5 | 4.8 | official | |
| 4 | claude-fable-5 | 6.5 | official | |
| 5 | claude-sonnet-5 | 6.7 | official | |
| 6 | gemini-3.7-flash | 9.2 | official | |
| 7 | muse-spark-1.2 | 24.2 | official | |
| 8 | gpt-5.6-sol | 27 | official | |
| 9 | gpt-5.6-terra | 37.3 | official | |
| 10 | gpt-5.6-luna | 50 | official | |
| 11 | grok-4.6 | 50.2 | official | |
| 12 | kimi-k3 | 52.7 | official |