Glossary term
Glossary term
Governance and Compliance
Design of when humans review, approve, override, monitor, or audit agent actions.
Anthropic's 'responsible scaling policy' mandates human-oversight design for high-capability models - defining specific oversight checkpoints, approval requirements, and emergency-shutdown procedures.
The EU AI Act's Article 14 requires natural persons overseeing high-risk AI systems to understand the system, intervene, and override outputs - a credit-decision platform was redesigned with a one-click override panel.
UiPath's AI governance framework defines oversight tiers: Tier-1 (fully automated), Tier-2 (sampled human review), Tier-3 (mandatory approval before action) - each tier has defined risk criteria and SLA requirements.