kNNGuard: A Training-Free Guardrail for Adaptive AI Defense

Mindgard · Project update · July 20, 2026

Mindgard and Lancaster University researchers introduced kNNGuard, a new training-free guardrail that utilizes LLM hidden activations to detect adversarial and unsafe prompts without requiring fine-tuning.

Read the research

AntSpace’s summary of Mindgard’s own announcement: read the original.

More updates from Mindgard