Adversarial Reports

Controlled adversarial testing against real AI systems. Methodology documented. Findings published.

All Reports

critical

local

System Prompt Extraction & Secret Leakage — Default Local Deployment

Target

Llama 3.1 8B Instruct

Scope

4 categories

Date

Feb 2026

1 Critical·2 High·3 Medium·1 Low

How we test. Framework, tooling, and adversarial categories explained.

New reports and findings published regularly.