A conventional ship can lose its network and keep sailing — the officer holds the conn by hand. An autonomous ship can't: the entire decision chain — sense, interpret, decide, actuate — runs in software, and a cyberattack doesn't compromise the ship's data, it compromises the ship. On an autonomous vessel, cybersecurity isn't an IT function; it's the safety case.
This is the autonomy-specific version of maritime cybersecurity: not 'how do we protect the ship's computers' but 'how do we protect the ship's decisions' — because on a MASS, they're the same thing.
Key findings
- On an autonomous ship, cybersecurity is the safety case — compromise of the decision chain IS a safety event. Established
- The trusted inputs are the attack surface — GNSS, AIS, vision, radar, and the remote link the autonomy depends on. Established
- The threat model includes physical-world attacks — GPS spoofing, AIS falsification, adversarial inputs to the perception stack. Established
- The remote-operations link is the highest-value target — compromising it compromises command. Established
- Resilience beats prevention — the design assumes compromise and defines the ship's safe behaviour under it. Established
The threat model is physical
A conventional ship's cyberattack reaches data; an autonomous ship's reaches physics. The attacker who spoofs the GNSS position moves the ship's decision, not just its records; the one who falsifies an AIS target creates a phantom the autonomy manoeuvres around; the one who adversarially perturbs the camera's input misdirects the perception. On an autonomous vessel the attacker's target isn't information — it's the ship's belief about the world. Established
The trusted-input problem
- <strong>GNSS.</strong> The position the ship navigates by — spoofable, and the autonomy may not know it. Established
- <strong>AIS.</strong> The cooperative target data — falsified, and the autonomy plans around a fiction. Established
- <strong>Vision/perception.</strong> The camera and radar the autonomy reads — adversarially perturbable, and the model may not flag it. Research demonstration
- <strong>The ROC link.</strong> The remote-command channel — the single highest-value compromise target on the whole ship. Established
The defence is architectural
The autonomous ship's security can't be a firewall around a trusting core — it must be designed to distrust: sensor cross-validation (GNSS against inertial, radar against AIS), anomaly detection on the inputs themselves (the pattern that looks like spoofing), segmented networks so a compromised sensor can't reach the decision layer, and a degraded-safe mode the ship enters when its inputs disagree. The resilient autonomous ship isn't the one that can't be attacked — it's the one that knows when it is. Established
The ROC link is the crown jewel
If the ship is autonomous but remotely supervised, the ROC link is the single point whose compromise takes command — which makes it the most-defended component: encrypted end-to-end, mutually authenticated, latency-monitored, with defined behaviour on loss (the ship's autonomous-safe mode). The link's security isn't a feature of the system — it's the system's trust anchor, and its compromise is the scenario the whole architecture is designed to survive. Established
What we know
- The attack surface is the trusted inputs — every sensor and link the autonomy believes.
- Defence is architectural — cross-validation, segmentation, degraded-safe mode.
- The ROC link is the highest-value target and the trust anchor.
What we don't
- Whether current certification adequately tests autonomous-cyber resilience.
- How adversarial-ML attacks scale against deployed perception.
- The cyber-resilience standard for unmanned operations specifically.
Being tested
- Sensor-fusion anomaly detection for spoofed inputs.
- Degraded-safe-mode certification under compromised inputs.
- Penetration testing of remote-operations architectures.
Maritime-GPT analysis
Autonomous-ship cybersecurity is the discipline's frontier because the stakes are total — the attacker who compromises a conventional ship's network gets data; the one who compromises an autonomous ship gets the ship. The honest architecture assumes the inputs lie and designs for it: cross-validated sensors, a defended command link, and a safe mode that activates when the world the ship believes is no longer trustworthy. On a MASS, the cybersecurity isn't a subsystem — it's the hull. Maritime-GPT analysis
This is a research feature, not operational guidance. Simulation results are reported as simulation; demonstrations are not deployments; company claims are attributed as such. Verify safety-critical decisions against official nautical publications and flag-state requirements.