Deep Seabed Security: The Infrastructure Behind Digital Power

Updated website edition · 8 October 2026. Based on the 67-page report dated 1 October 2026.

A submarine cable is secure only to the extent that the services depending on it can continue through a credible disruption. The useful boundary includes the seabed asset, landing facilities, power, terrestrial links, management software, suppliers and the ability to restore service. Counting cables or commissioning a surveillance platform does not answer that continuity question.

The original report’s strongest judgement is that investment should reduce shared dependencies and shorten recovery before assuming every outage can be prevented. This edition retains its engineering, cyber, sensing, India, repair and quantitative frameworks. The full research report remains an original source; this is its public website synthesis.

The current evidence changes the emphasis

ITU’s current backgrounder records more than 170 cable repairs worldwide in 2025, citing ICPC. It identifies accidental fishing and anchoring, then natural hazards, among the principal vulnerabilities. This is a dated global repair count, not 170 hostile attacks. ITU backgrounder.

On 10 July 2026 the international advisory body approved recommendations covering deployment and repair, risk monitoring and geographical diversity. That shifts the implementation test from another declaration of concern to demonstrable restoration capability. ITU announcement.

The source report’s emphasis on ordinary failure is therefore compatible with a serious security assessment. Deliberate interference remains a possibility requiring investigation; it is not the default explanation for every loss of connectivity. Prevention, operational recovery and attribution run on different clocks.

Follow the complete service path

A customer uses an application, not a fibre pair. An apparently independent cable may still depend on the same landing facility, inland duct, authentication service or network-management provider. A physically intact route can be unusable if the customer lacks a capacity entitlement, if onward connectivity is congested, or if the application cannot tolerate the available latency.

This approach produces a different procurement decision from simply buying more bandwidth. A small independent path may contribute more to continuity than a large path sharing the dominant failure. Conversely, the smaller link may preserve essential transactions while being unable to support ordinary peak traffic. Both statements can be true.

Figure 01 · Conceptual framework

A cable is one part of the service

How the parts connect

1. AMarine linklands atShore facilities
2. BShore facilitiesconnects throughInland network
3. CInland networksupportsEssential service
4. DControl systemsadministersShore facilities
5. EControl systemsadministersInland network

India’s numbers need dates and units

The December 2024 Ministry of Communications release described around 17 international cables and 14 distinct landing stations. Its 138.606 Tbps lit and 111.111 Tbps activated figures refer to end-2022. These are historical baselines, not October 2026 system counts or peak traffic measurements. Official release.

A March 2026 parliamentary answer reported industry-sourced operational data-centre capacity of 790 MW in Mumbai/Navi Mumbai and 305 MW in Chennai, within a table totalling 1,520 MW. Together they represent about 72.0% of that table’s power capacity. This is not 72% of India’s submarine bandwidth. The same release listed IEX, SEA-ME-WE 6, 2Africa and Raman as under commissioning. PIB, 18 March 2026.

This refresh does not turn those March statuses into an October operational inventory. Landfall, physical completion, acceptance and available commercial capacity are different milestones. Where a later end-to-end status has not been independently established, a national continuity model must require current operator confirmation before treating announced capacity as a usable fallback.

The public-policy implication is to measure surviving service, not celebrate gross additions alone. Publishing aggregate readiness indicators can support accountability without presenting precise facility layouts or live vulnerability maps.

Test common failure before averaging availability

The original report provides a useful, explicitly hypothetical capacity model. Three links carry 50, 30 and 40 Gbps. The first two share a regional exposure. Normal demand is 80 Gbps, essential demand is 30 Gbps, and the planning policy permits use of 80% of surviving contracted capacity. The headroom policy is an assumption, not a universal industry standard.

Essential demand fits in all three cases only if traffic priorities and downstream dependencies also work. A nominal 50% surplus over normal demand is therefore not proof of resilience. The critical assumption is independence under the event being tested.

A second model holds each path’s snapshot unavailability at 2%. With failure-indicator correlation rho, joint unavailability equals $p^2+\rho p(1-p)$. At rho = 0, the result is 0.040%; at 0.1 it is 0.236%; at 0.5 it is 1.020%; at 0.9 it is 1.804%. These are illustrative sensitivity calculations, not measured cable probabilities. They cannot be converted directly into annual incident counts.

A real model should remove all nodes and links affected by a named common event and then test service-specific demand, entitlement, latency and congestion. Adding nominal spare capacities without checking the same downstream bottleneck can count the same protection twice.

Figure 02 · Qualitative comparison

What makes restoration possible

What makes restoration possible

Capability	Evidence to request
Access	Evidence to requestApplicable permits and response responsibilities
Resources	Evidence to requestAvailability of parts and qualified repair capacity
Fallback	Evidence to requestUsable entitlements and endpoint compatibility
Coordination	Evidence to requestPractised escalation and ownership
Validation	Evidence to requestEvidence that essential services recover

Cyber recovery is part of physical resilience

The wet plant can be intact while operating systems, privileged accounts or supplier access fail. A credible exercise should therefore include trusted out-of-band administration, known-good configurations, access revocation, restoration of logging and verification before ordinary service resumes. These are defensive design requirements, not claims that every cable system shares an identical architecture.

Recovery needs an owner across commercial boundaries. A carrier may see its segment as healthy while the customer remains unable to authenticate or complete a transaction. Contracts should identify the lead responder, required evidence, supplier cooperation and the point at which the service is considered restored.

Restored connectivity is not the same as recovered trust. If the initiating event may involve compromise, routing traffic through a surviving system without verifying its management and identity controls can propagate the problem.

Sensing adds evidence; it does not prove intent

The source examines distributed acoustic sensing, polarization, interferometry and purpose-built scientific sensors. They measure different physical effects and have different spatial resolution, installation and calibration requirements. A long-range environmental detection should not be sold as a universal small-target identification capability.

New primary research makes old blanket limits unreliable. A 2026 EGU abstract reports multispan DAS observations over about 4,400 km with approximately 100 m spatial resolution. This is a research demonstration, not an installed capability guarantee for every commercial cable. Research abstract.

A National Physical Laboratory update on 22 September describes work toward tsunami detection using submarine cable interferometry, with sensitivity and automated analysis still part of the development task. NPL update.

The operational sequence should keep detection, location, classification and attribution distinct. Correlating multiple feeds derived from the same underlying observation does not create independent corroboration. A nearby vessel track alone does not establish seabed contact or intention.

Figure 03 · Conceptual framework

Test the loss that matters

How the parts connect

1. AMap shared exposurescenarioRemove a dependency
2. BRemove a dependencysurviving pathsCheck usable fallback
3. CCheck usable fallbackapplication testTest essential tasks
4. DTest essential tasksdecisionRevise the investment
5. ERevise the investmentreviewMap shared exposure

Restoration is a system, not a vessel count

Useful restoration time includes fault identification, permits, safe access, vessel mobilisation, compatible spares, weather, repair, testing and return to service. A repair fleet can look adequate in aggregate while its available vessels are committed elsewhere or lack permission to enter the required jurisdiction.

Measure the distribution of time to restore essential service under simultaneous regional demand. Separate temporary rerouting from physical repair and full normal capacity. Reducing administrative delay may improve service more quickly than adding a sensor that does not change the response.

A resilience programme should therefore maintain a current asset-and-dependency register, verify contractually usable fallback capacity, exercise joint recovery, confirm repair arrangements and review bottlenecks after each material network change. The public dashboard should report outcomes and unresolved gaps, not imply that every approved project is operational.

What should change the assessment

Three kinds of new evidence matter most: independent capacity actually accepted into service; demonstrated reduction in recovery time during realistic exercises; and calibrated sensing that improves a decision at an acceptable false-alert rate. Another cable announcement or a larger surveillance dataset may help, but neither establishes these outcomes by itself.

The central policy choice is between paying for visible assets and paying for a demonstrably recoverable service. The full source research shows why the second is the more demanding and useful standard.