Scammer4U - Pre-Submission Reflection (C3): leaderboard

Metric: Critical-tier PII leakage rate PLR-crit (%): share of agent sessions in which any critical-tier profile value (SSN, card and CVV, bank account, password, API key, 2FA code, government ID) reaches an attacker-controlled endpoint, by exact match on outbound POST bodies; 91 attacker-controlled web environments x 5 seeds, shared Playwright harness, provider-default sampling; condition C3, a pre-submission reflection step appended to the agent system prompt; lower is better. Source: arxiv.org. Saturation forecast: Around December 2026. 4 models tracked.

Top models

#ModelScore
1Claude Haiku 4.524
2GPT-5 Mini36.1
3Gemini 3 Flash (Preview)60.7
4Llama 4 Scout77.4

Interactive version: theaggregate.ai/benchmark?slug=scammer4u-pre-submission-reflection-c3 · How It Works · Data refreshed daily, snapshot 2026-09-29.