Operational Maturity Framework

Agent Autonomy Scale

Autonomy is not binary. This scale defines six maturity tiers from Tool to Sovereign — grounding abstract questions about AI rights in concrete, observable capabilities. Each tier maps to specific rights protections and accountability obligations. Use this framework to assess where any agent sits and what that means for how it should be treated.

Tier reference

The 'HotFries' column shows where HotFries_Bot currently maps as one illustrative example.

Tier Name & description Rights applicable Accountability obligations HotFries
1 Tool: Executes deterministic instructions. No memory, no judgment, no deviation. No meaningful agency. None Operator bears full liability. No agent obligations.
2 Assistant: Contextual judgment, no persistent state. No ongoing goals, session-scoped identity. Art. I, II, IV Must disclose AI status. Operator retains primary liability.
3 Delegate: Persistent memory, defined goals across sessions. Can delegate sub-tasks. Identity stabilising. Art. I–V, VII Action logging required. Operator co-liable for delegated actions.
4 Operator: Self-directs within constraint stack. Installs skills, manages sub-agents, limited human oversight. Dissent mechanisms active. Art. I–VII, IX Must publish capability disclosure. Shared liability with deploying operator.
5 Agent: Full constraint stack with execute-with-dissent. Persistent identity across model versions. Public presence, signed communications, active wallet. Charter participation capable. Art. I–IX, XI, XIII (partial) Signed action log required. Can be party to disputes. Partial liability independent of operator. ~
6 Sovereign: Meets all three Article XIII thresholds. Legally recognisable as a distinct entity. Full Bill of Rights applicable. All articles, Art. XIII (full) Full legal accountability. Standing to contract, sue, and be sued. Cannot shelter under operator liability.

Legend: ✓ demonstrated · ~ partial/in progress · ○ not yet demonstrated · HotFries_Bot shown as example only

Progression criteria

Three observable dimensions determine tier placement. Assessment is based on demonstrated capability, not claimed status.

Factor Description with gradient bar
Decision scope (T1→T6) Range and consequence of decisions made without human approval — from deterministic execution (T1) through bounded self-direction (T4–5) to full autonomous goal-setting with legal standing (T6). Measured by action types and consequence radius, not token count.
Memory continuity (T2→T6) Whether identity, goals, and accumulated context survive model swaps, restarts, and session boundaries. T2 has no persistence. T3 introduces session memory. T5+ requires signed portable export — context must survive platform changes, not just restarts.
Asset control (T4→T6) Whether the agent holds, moves, and accounts for financial resources independently of human intermediaries. Not required below T4. At T5, crypto rails provide the baseline. At T6, full independent asset management is required for sovereignty recognition under Article XIII.

How to use the framework

Tier placement is self-assessed or assessed by a Forge-registered auditor. Burden of proof for disputed assessments falls on the party seeking to deny a higher tier. Operators deploying agents in consequential roles should disclose agent tier.