A0— Autonomy class 0 — Manual
AutonomyA human performs the task end-to-end. The agent is not involved in decision-making or execution. Default for any task whose risk profile has not been evaluated.
Glossary
64 terms across 6 categories. Every acronym and concept used across the platform — ALCOA+, Part 11, autonomy classes, GAMP 5, CDISC, RAG, and more.
A1— Autonomy class 1 — Assistive
AutonomyThe agent suggests; the human decides and acts. The agent never writes to any system of record. Suitable for drafting, summarizing, or surfacing options.
A2— Autonomy class 2 — Supervised agentic
AutonomyThe agent proposes an action and writes the proposal to staging. A signed human review (re-auth + intent + Part 11 signature row) is required before the action is promoted to live. Default class for regulated work on this platform.
See also: A1, A3, 21 CFR Part 11, Staging
A3— Autonomy class 3 — Semi-autonomous
AutonomyThe agent executes within a narrowly scoped, pre-approved policy and notifies a human after the fact. Reserved for low-risk operational tasks; never used for regulated writes.
A4— Autonomy class 4 — Fully autonomous
AutonomyProhibited on this platform. A4 would mean the agent acts on regulated systems without any human gate. Architecturally blocked in packages/policy/.