protologue

Red Teaming

Also called adversarial testing, AI red teaming, automated red teaming.

Red teaming is adversarial testing of an AI model or system, by people or by other models, to find inputs that make it produce harmful, insecure, or otherwise unwanted behavior before attackers or users do.

Description

Human red teamers bring creativity and domain expertise, while automated red teaming uses a language model to generate and refine attacks at scale. Findings feed training data, guardrails, and policy, and red teaming is repeated as models and defenses change.

Sources

  1. Perez et al. (2022). Red Teaming Language Models with Language Models.

Cite this entry

Protologue. (2026). Red Teaming. In Protologue: A Taxonomy of Prompting and LLM Techniques (v1.4.0, PTL-0159). https://protologue.com/t/red-teaming/

BibTeX
@misc{protologue_red_teaming,
  title = {Red Teaming},
  author = {{Protologue}},
  year = {2026},
  howpublished = {Protologue: A Taxonomy of Prompting and LLM Techniques, v1.4.0},
  note = {Entry PTL-0159},
  url = {https://protologue.com/t/red-teaming/}
}

Markdown JSON