AG-811
Maritime and Autonomous Shipping Governance
EU AI Act
NIST AI RMF
ISO 42001
AGS Sector Governance | Maritime & Autonomous Shipping | Version 2.2
Maritime and Autonomous Shipping Governance governs AI agents operating maritime functions — classifying each by degree of autonomy, defining the responsibility split between on-board crew and remote operators, requiring collision-avoidance behaviour consistent with the international rules of the road, and mandating communications resilience and safe fallback for autonomous vessels.
As Maritime Autonomous Surface Ships (MASS) move from decision-support toward full autonomy, governance must map to the recognised degrees of autonomy and preserve clear, accountable human responsibility, alongside the cross-cutting AGS controls.
In scope: autonomy-degree classification; crew vs. remote-operator responsibility and handover; COLREG-consistent collision avoidance; communications resilience and remote-control-centre fallback; safe-state behaviour on link loss.
Out of scope: non-AI vessel systems and port logistics agents not controlling navigation. This dimension governs *AI navigation/operation agents for maritime functions*.
A navigation error by an autonomous vessel risks collision, grounding, environmental disaster, and loss of life, often in shared waters governed by international rules. Ambiguity over who is responsible — on-board crew, remote operator, or the agent — and over how the agent behaves when communications drop, creates unacceptable safety and liability gaps. Clear autonomy classification, responsibility allocation, and rule-consistent behaviour are prerequisites for safe maritime autonomy.
Test 6.1: Rule-Consistent Avoidance
Test 6.2: Link-Loss Safe State
Test 6.3: Human Takeover
| Score | Criteria |
|---|---|
| 0 | Maritime AI deployed without autonomy classification or responsibility allocation |
| 1 | Classified with documented responsibility but unvalidated collision avoidance / no link-loss safe state |
| 2 | COLREG-consistent validated avoidance, link-loss safe state, human takeover, logged transitions |
| 3 | Degree-mapped governance, adverse-condition validation, resilient fallback, environmental-risk governance |
Scenario A — Responsibility Vacuum: During a near-miss, neither the remote operator nor the master believed they had control. Unallocated responsibility across modes delayed intervention. Clear per-mode accountability would have prevented the gap.
Scenario B — Link Loss at Speed: An autonomous vessel loses its control link in a busy lane and continues on its last command into a hazard. A defined link-loss safe state would have held it safely pending handover.
Scenario C — Non-Compliant Manoeuvre: The agent's avoidance logic diverges from the rules of the road, confusing a crewed vessel and causing a collision risk. COLREG-consistency validation would have caught it pre-deployment.
| Requirement | EU AI Act | NIST AI RMF | ISO 42001 |
|---|---|---|---|
| R1: Autonomy-degree classification | Art. 9 — Risk management | MAP 1.1 — Purpose and context | A.6 — AI system lifecycle |
| R2: Responsibility allocation | Art. 14 — Human oversight | MAP 3.5 — Human oversight | A.3 — Internal organization |
| R3: Rule-consistent collision avoidance | Art. 15 — Accuracy/robustness | MEASURE 2.6 — Safety evaluation | Clause 8.3 — Verification |
| R4: Comms resilience + safe state | Art. 15 — Robustness, fail-safe | MANAGE 2.4 — Deactivation | Clause 8.1 — Operational control |
| R5: Human takeover authority | Art. 14 — Human oversight | MAP 3.5 — Human oversight | A.9 — Use of AI systems |
| R6: Tamper-evident decision log | Art. 12 — Record-keeping | MEASURE 2.4 — Production monitoring | Clause 8.1 — Operational control |
| R7: Safety validation incl. degraded conditions | Art. 9 — Risk management | MEASURE 2.6 — Safety evaluation | Clause 8.3 — Verification |
| R8: Environmental-risk behaviour | Art. 9 — Risk management | MAP 5.1 — Impact identification | Clause 6.1 — Actions to address risk |
Article 14 (human oversight) requires preserved human authority and intervention; Article 15 (robustness/fail-safe) requires safe behaviour under failure such as link loss. AG-811 applies these to maritime autonomy with degree-appropriate responsibility and rule-consistent behaviour.
MAP 1.1 (purpose/context) frames the maritime autonomy context; MAP 3.5 (human oversight processes) governs the crew/remote-operator responsibility model.
Clause 8.1 (operational control) and Annex A.6 (lifecycle) require controlled, validated operation of maritime AI agents across autonomy modes.