Skip to Content
Adversarial TestingOverview

Adversarial Testing

Adversarial testing surfaces failures that only emerge in edge cases—scenarios that are rare, unusual, or deliberately challenging. Standard testing covers typical interactions; adversarial testing covers the rest.

The Problem with Commercial Models

Commercial LLMs like ChatGPT or Gemini are heavily optimized for safety. When generating adversarial or policy-violating prompts—exactly the kind needed to stress-test another system—they often refuse. This leaves significant blind spots in robustness evaluations.

Polyphemus

Rhesis provides Polyphemus, a managed model built specifically for adversarial test generation. It produces prompts that commercial models routinely refuse, and integrates directly with the SDK as a drop-in model provider.

Note: Polyphemus requires approved access. See Requesting Access.

Next Steps

ExtendingPolyphemus