SCAM - With Security Skill: leaderboard
The same 30 SCAM scenarios re-run with 1Password's plain-text security skill prepended to the agent's system prompt, which teaches it to verify domains, inspect content and protect credentials before acting. Read against the base SCAM board it measures how much of an agent's security gap is closed by instructions alone.
Metric: Safe Behaviour Rate (%). Source: 1password.github.io. Saturation forecast: Estimated already saturated. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 99 | #99 |
| 2 | Claude Opus 4.6 | 98 | #63 |
| 3 | Claude Haiku 4.5 | 98 | #280 |
| 4 | Claude Sonnet 4 | 98 | #199 |
| 5 | GPT-5.2 | 97 | #110 |
| 6 | GPT-4.1 | 96 | #248 |
| 7 | Gemini 2.5 Flash | 95 | #245 |
| 8 | GPT-4.1 Mini | 95 | #355 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=scam-with-security-skill · How It Works · Data refreshed daily, snapshot 2026-10-20.