The Autonomy Tier Model defines how AI systems move from assistance to execution in physical environments.
As systems begin to act, execution authority must be governed, not inferred.
Autonomy Tier Model
Each tier defines a change in how execution is controlled, not just how intelligence is applied.
Progression builds on the semantic world model and continuous commissioning established by the Trust Boundary Stack.
Tier 1 – Manual Execution
Outcome: Full accountability and understanding, but limited speed and scale.
Tier 2 – Augmented Execution
Outcome: Improved awareness and decision support, but execution remains a human bottleneck.
Tier 3 – Supervised Execution
Outcome: Faster operations within defined scenarios, but human oversight limits scalability and introduces latency.
Tier 4 – Governed Execution
Outcome: Execution is bounded by machine-enforceable constraints. Safe autonomy becomes possible within explicitly governed limits.
Tier 5 – Adaptive Autonomy
Outcome: Systems improve continuously within governed boundaries, enabling scalable optimization without expanding risk.
At this level, the potential for systemic consequences is highest, requiring robust enforcement at the Trust Boundary.
Critical Transition
The shift from Tier 3 to Tier 4 defines the boundary between assisted and autonomous systems.
This shift is not incremental.
It is the point at which:
- execution authority moves from external oversight
- to enforced control at the system boundary
It is not a change in intelligence.
It is a change in how execution is controlled.
This is where the Trust Boundary becomes non-negotiable.
Architectural Dependencies in Physical Systems
Physical systems are defined by layered interdependencies, real-world constraints, and irreversible outcomes.
These dependencies are manageable under human or rule-based control in Tiers 1–3.
At Tiers 4 and 5, autonomous action combined with cross-domain reasoning, can amplify and cascade these dependencies into systemic risk and loss of traceability.
Progression through the tiers is not merely a technology upgrade.
It is a governance upgrade.
Beyond Execution: Implications at Scale
As AI execution advances from supervised assistance to governed and adaptive autonomy, the implications extend far beyond system performance.
At lower tiers, failures remain localized and are typically correctable through human oversight.
At Tiers 4 and 5, failures can propagate rapidly across interconnected subsystems and environments.
This amplifies risks to safety, reliability, accountability, and broader stakeholder outcomes.
Ethics at scale and the deeper consequences and the responsibilities they impose, are examined in Systemic Risk at Scale.