Search the Atlas

Search risks, controls, and glossary terms

MediumNo ProvenanceDemonstrated

Missing content provenance / watermarking

Transparency & Explainability

Description

AI-generated content is indistinguishable from human content, enabling misuse and eroding trust.

Example scenario

AI-drafted customer communications are sent without any indication they were machine-generated.

Real-world evidenceDemonstrated

Regulatory frameworks requiring model documentation (EU AI Act, SR 11-7, EBA guidelines) are now in force, and audits routinely find gaps, but large-scale enforcement actions specifically citing absent model cards or technical documentation as a primary violation are still emerging rather than a confirmed pattern of production incidents.

Primary mitigations

  • Watermarking/C2PA provenance
  • AI-content labelling
  • disclosure on generated material.

Detection signals

Provenance-coverage metric; label-presence checks.

Mitigating controls

4
Non-agentic controls

Related risks in Transparency & Explainability