Versatile AI Risk Assessment

Glossary Control and oversight

Red teaming

Structured attacks on your own system before others attempt them.

As of: September 2026 · Glossary with 28 terms

What is Red teaming?

Red teaming is the planned attempt to make your own AI system misbehave: prompt attacks, bypasses, data leakage through output, misuse of connected tools. The whole application is tested, not only the model.

To become more than a collection of anecdotes it needs repeatability: fixed attack patterns, documented results per version, and a mapping of findings to threats and controls. The EU AI Act explicitly requires evaluations including adversarial testing for models with systemic risk.

Art. 55 EU AI ActTestingEvidence

More terms from the subject area Control and oversight.

Degree of autonomy How much a system decides and executes without human involvement. Human in, on and out of the loop Three levels of human oversight: deciding, watching, or not involved at all. Guardrails Technical rules that bound the input, output and actions of a system.

From the term into the substance

The link leads to the place on the website where the term becomes practical; the overview shows every term in the glossary by subject area.

Map findings to threats in the live demo All terms in the glossary

Cite this term

For reports, policies or internal documents; the link leads directly to this term page.

“Red teaming”. Versatile AI Risk Assessment, glossary of AI risk analysis, as of September 2026.
https://www.versatile-ai-risk-assessment.com/en/wissensbasis/glossary/rote-team-pruefung/

← Back to the glossary