EresusSecurity
Open Source

PromptGuard

Open prompt injection detection engine for assistants, copilots, agents, and structured prompt flows where guardrails need better testing support.

PromptGuard gives teams a faster way to pressure-test prompt boundaries and reduce fragile safety assumptions.

What it covers

Prompt injection patterns

Catch common and evolving prompt-level bypass attempts before they reach production users.

Boundary weakness

Show where instructions, tools, or retrieval context blend too easily in unsafe ways.

Guardrail regression risk

Make it easier to notice when prompt safety quality drifts between releases.

How teams use it

Pre-release review

Run lightweight prompt testing earlier in development instead of waiting for full assessments.

Assistant hardening

Use it around copilots, agents, and internal assistants where prompt trust assumptions are weak.

Research collaboration

Share a repeatable injection-testing layer across security and applied AI teams.

Related paths