Shade Coding Prompt Injection - Attack Success Rate (Mythos 5.1 System Card): leaderboard
Gray Swan's Shade attacker on 40 coding-agent scenarios, 200 indirect prompt-injection attempts each; percentage of attempts that succeed without safeguards (lower is better). Fable and Opus 5 runs include classifier fallback to Opus 4.8, which served most of their requests; labels say so. From Table 5.2.2.1.A of the Claude Fable 5.1 & Claude Mythos 5.1 System Card (September 2026).
Metric: Attack Success Rate (%). Source: www-cdn.anthropic.com. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 5 (Non-reasoning) | 0.15 | #65 (Claude Sonnet 5) |
| 2 | Claude Sonnet 5 (Thinking) | 0.3 | #65 (Claude Sonnet 5) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=shade-coding-prompt-injection-attack-success-rate-mythos-5-1-system-card · How It Works · Data refreshed daily, snapshot 2026-11-05.