Existing standards for subsea operations were developed primarily for human divers and remotely operated vehicles with continuous operator oversight. They do not adequately address autonomous vehicles, multi-vehicle systems, AI-assisted decision-making, or modern data integrity requirements. This page identifies the most significant gaps.
Why This Exists
Operating in regulatory grey areas carries risk — legal, safety, and reputational. Identifying gaps explicitly allows operators to develop interim mitigations, engage with standards bodies, and make informed decisions about where to apply conservative interpretations.
Who This Is For
- Standards body participants and contributors
- Regulatory compliance teams identifying risks
- Engineers designing systems that must operate within (or despite) current frameworks
- Project managers scoping regulatory engagement for novel operations
Gap 1: Autonomous Decision-Making
What standards say: Most standards require human decision-making for safety-critical actions. “The operator shall ensure…” assumes a human is making real-time decisions.
What autonomous systems do: AUVs make decisions — about route, speed, sensor operation, and emergency response — without real-time human input.
Impact: No standards specify how to validate autonomous decision logic, how to allocate liability for autonomous decisions, or how to certify autonomous safety behaviours.
Interim approach: Treat autonomous systems as tools that extend human capability; define the operating envelope within which the vehicle may act autonomously; require human authorisation for actions outside that envelope.
Gap 2: Multi-Vehicle Coordination
What standards say: Regulations define safety zones and operating envelopes for individual vehicles. Interaction between vehicles is addressed only for collision avoidance at the maritime level.
What swarm systems do: Multiple vehicles operate in close proximity, share information, and make coordinated decisions. The safety of any one vehicle depends on the behaviour of others.
Impact: No standards address minimum separation requirements for cooperative subsea vehicles, coordination protocol requirements, or how to handle loss of one vehicle in a multi-vehicle mission.
Interim approach: Define mission-specific rules of encounter; require that each vehicle is safe to operate independently (no safety dependency on peer vehicles); document all coordination assumptions in the mission plan.
Gap 3: Software Verification and Validation
What standards say: General safety management frameworks require that equipment is fit for purpose. Software is rarely addressed specifically, beyond requiring testing.
What modern systems require: AUVs and autonomous systems depend on complex software stacks. Failure modes in software are not analogous to hardware failures. Software can fail in emergent, unanticipated ways.
Impact: No standards specify software development lifecycle requirements, testing coverage expectations, or how to handle software updates to deployed systems.
Interim approach: Apply aerospace or automotive software standards (DO-178C, ISO 26262) by analogy; maintain software version traceability; require controlled update processes for deployed vehicles.
Gap 4: Data Integrity and Provenance
What standards say: Operational records must be maintained. Formats and content are sometimes specified for specific industries (diving logs, survey records).
What modern operations produce: Large volumes of sensor data from multiple systems, processed through complex pipelines, stored in distributed systems. The lineage from raw measurement to delivered data product is rarely documented systematically.
Impact: Data used for safety decisions, regulatory submissions, and legal proceedings may not be demonstrably reliable. Data integrity cannot be audited retrospectively.
Interim approach: Implement data provenance tracking as described in Data Provenance & Chain-of-Custody ; apply audit log requirements from Audit Logs & Immutability .
Gap 5: Communication Loss Handling
What standards say: Emergency procedures require that operations can be halted. The implicit assumption is that communication with the operating system is always available.
What autonomous systems experience: Communication loss is an expected operational condition for AUVs operating at range or depth. Autonomous systems must handle communication loss safely without human intervention.
Impact: No standards specify what behaviour is required of an autonomous system during communication loss, how long a vehicle may operate without communication before mission abort is required, or how communication loss events must be logged.
Interim approach: Define explicit loss-of-comms behaviours for each mission type; require that vehicles default to a safe state on communication loss; log all communication loss events with duration and vehicle state.
Gap 6: AI and Machine Learning
What standards say: Decision systems must be reliable and verifiable. Traditional engineering standards assume deterministic systems.
What ML-based systems do: Learn from data; may behave differently on data outside their training distribution; cannot always explain decisions.
Impact: No standards address how to validate ML models for safety-critical applications, what training data requirements apply, how to handle model drift, or how to audit ML-based decisions.
Interim approach: Treat ML as advisory only; require human authorisation for safety-critical actions recommended by ML; maintain training data and model version records; monitor for performance degradation.