The Royal Navy (RN) has all but declared its hand that autonomous surface systems will form a significant part of its fleet as it moves into the Navy after next. The Maritime Operating Concept (MarOpC) signposts autonomy five times. The RN needs its ships to deliver credible military effect, autonomy can help increase mass, reduce crewing and complete tasks ultimately with little or no oversight of most systems. Small boats such as the Autonomous Pacific 24 (APAC24) are going to form part of the answer increasing capability for larger more conventional platforms but for the RN to really enjoy the benefits of autonomy they’re going to have to move away from remote control IMO Degree 3 and into the world of ship autonomy in IMO Degree 4. If this is possible it is a win for defence but how do we get there? Given we’re still grappling with the challenges of small boats and degree 3. Historically naval ships are built to naval codes which stem from their commercial equivalents adding in the military delta. With nothing expected from the International Maritime Organisation (IMO) in terms of legislation until at least 2025 and the Maritime and Coastguard Agency have only just released a workboat code annex to support the regulation of remote operated systems less than 24 meters. The RN don’t have the luxury of well defined regulation and almost by accident have become the front line in the regulation space placing pressure on getting it right. These highly autonomous platforms have a significant amount invested in the sensor suite and ultimately the decision engine likely to be AI based so how can you regulate these? Synthetic environments might be part of the answer and this paper looks where synthetics have been used before to support regulatory outputs, the limitations of the current methods already implemented and how these might be optimised to install trust in the systems before going to sea. We already accept computer modelling for regulatory purposes in other areas however, never with the stakes so high. There are many elements that go into the validation of the synthetic model before this is even able to be useful to validate the autonomy system itself- this paper explores these and how the challenge might be tackled. It will also explore some of the interdependencies that will be required to make the synthetic testing credible.