Summary

  • On 14 March 2024, regulators, operators, and independent measurement systems recorded severe Internet disruption across western, central, and southern Africa as ACE, MainOne, SAT-3, and WACS experienced faults. The public record does not establish one exact break time or sequence for all four systems. [1][8][11]
  • Internet Society identified 13 affected countries, but one figure caption in its later report refers to 13 March while the narrative record from regulators, operators, and measurements centers on 14 March. That discrepancy should remain visible rather than being replaced with a manufactured universal timestamp. [9]
  • Ghana's regulator estimated capacity losses of 90 percent for ACE, about 90 percent for MainOne, 100 percent for SAT-3, and 90 percent for WACS, totaling 1,596.6 Gbps for Ghana. Those Ghana-specific estimates are not a complete regional loss total. [2]
  • A later Ghanaian notice said all four landing providers had been cut off from international data services. That service-state statement concerns a different operational layer and moment from the earlier cable-capacity estimates, so the two records should not be flattened into a single percentage. [3]
  • Continuity varied because some networks could reach unaffected capacity. Documented examples included Moov Africa Ivory Coast using the differently routed Maroc Telecom West Africa cable, reported traffic shifts toward Equiano, Angola Cables redirecting traffic over SACS, and Bayobab activating reroutes and additional circuits. [10][12][13][15]
  • Cloudflare observed part of MainOne's AS37282 IPv4 space cease being announced for several hours, alongside loss and return of MainOne-hosted services. That is important reachability evidence, but it is not a complete reconstruction of every BGP update, policy decision, or alternate circuit used during the event. [11]
  • Local exchanges and caches reduced some dependence on damaged international capacity. Ghana's regulator cited content available through the Ghana Internet Exchange and later convened GIX, the Accra Internet Exchange, financial institutions, utilities, operators, and cable providers as constrained bandwidth affected essential digital transactions. [3][5]
  • Traffic restoration and physical wet-plant repair ran on different clocks. Alternate capacity and rerouting restored some reachability before repairs concluded on 6 April for SAT-3, 17 April for ACE, 29 April for WACS, and 8 May for MainOne. [7][14]
  • Ghana's final account attributed to cable providers the finding that all four faults occurred at crossings of the Le Trou Sans Fond submarine canyon off Abidjan. This supports a shared-corridor risk finding, but not negligence, sabotage, a single proven trigger, or a complete forensic account. [7][9]
  • Accountability follows practical control and retained evidence. Cable operators, carriers, networks, exchanges, content providers, regulators, and critical institutions controlled different parts of route design, capacity, routing, locality, restoration, and repair. Distributed control is not distributed blame.

Four cable names, one regional failure domain

A network diagram can display four submarine cable names and still conceal one consequential failure domain. ACE, MainOne, SAT-3, and WACS were separate systems with different ownership and operating arrangements, but their March 2024 faults removed substantial international capacity from countries that depended on the affected West African corridor. The relevant resilience question is therefore not how many colored lines appeared on a map.

It is whether the lines were independent enough, connected to usable terrestrial and international routes, commercially available to affected networks, accepted by routing policy, and sufficiently provisioned to carry diverted traffic.

The incident became visible as a regional connectivity crisis on 14 March. Ghana's National Communications Authority reported disruption to mobile and fixed data services and relayed preliminary accounts concerning faults affecting international cable infrastructure. Independent measurement systems also detected substantial outages across multiple countries, while public reporting described failures that disrupted access far beyond a single landing station. [1][8][16] None of those records, considered alone, identifies the precise physical sequence of every cable fault.

Together, however, they establish a multi-country event in which several important systems became unavailable or severely impaired within a narrow period.

Internet Society's after-event report lists Benin, Burkina Faso, Cameroon, Côte d'Ivoire, The Gambia, Ghana, Guinea, Liberia, Namibia, Niger, Nigeria, South Africa, and Togo as affected countries. [9] That set spans coastal landing markets, inland networks dependent on cross-border terrestrial connections, and countries in southern Africa. It demonstrates the reach of the failure domain, but it should not be treated as a league table of national preparedness. Each country had a different combination of cable access, terrestrial extensions, upstream providers, exchange participation, caches, traffic patterns, and emergency capacity.

Those differences produced materially different outcomes. Some networks lost broad international reachability, some retained partial access, and some recovered traffic through alternative systems before the damaged cables were repaired. A region-wide account that reports only the cable cuts misses the controls that determined those outcomes. Conversely, an account focused only on restored traffic misses the physical common-mode exposure and the long repair period. Both layers matter: wet-plant damage created the capacity shock, while routing, transit, interconnection, and commercial arrangements shaped the service experienced by users.

This distinction is central to infrastructure accountability. Installed infrastructure is an input, not proof of continuity. A second cable cannot protect a network if it follows the same vulnerable corridor, terminates in the same constrained dependencies, cannot be purchased during an emergency, lacks an active circuit, is unreachable under current routing policy, or has too little spare capacity. An exchange point cannot replace international transit, but it can keep eligible local traffic local. A reroute can restore reachability without repairing the damaged cable, but only if the route remains stable and usable under load.

The March event supplied concrete evidence for each of those boundaries. It showed common physical exposure near Abidjan, observable changes in route announcements, use of unaffected submarine systems, reliance on cross-border and local interconnection, constrained replacement bandwidth, and a repair campaign involving multiple cable owners and two vessels. The accountability test is therefore operational: what continued to work, what had to be activated, where capacity ran short, how restoration was measured, and which records can demonstrate that resilience claims match actual behavior.

A chronology with several clocks, not one outage percentage

The public record is strongest when its distinct clocks are kept separate. Internet Society's later report contains a figure caption referring to 13 March, while its surrounding narrative and the principal regulator, operator, and measurement accounts center on 14 March. [9] That may reflect a caption error, a time-zone or dataset boundary, or an observation not fully explained in the published material. The available evidence does not support choosing one of those possibilities as fact.

The responsible treatment is to disclose the discrepancy and use 14 March for the main event chronology while declining to claim one exact start time for every physical fault.

Cloudflare observed disruptions beginning around 05:00 UTC in The Gambia, Guinea, and Liberia. Its traffic data showed effects in Côte d'Ivoire and Ghana around 07:30 UTC, followed by disruption in other countries. [11] This sequence describes when connectivity changes became visible in Cloudflare's measurements. It does not reveal the moment each cable was damaged, the order of the physical faults, or when every provider recognized and declared an incident. Traffic measurements are downstream observations of a network event, not seabed instrumentation.

Ghana's second update provided an early, country-specific capacity view. The NCA reported estimated losses of 90 percent on ACE, approximately 90 percent on MainOne, 100 percent on SAT-3, and 90 percent on WACS. It calculated an aggregate loss of 1,596.6 Gbps for Ghana. [2] These figures are useful because they preserve cable-level distinctions, but they are estimates from an evolving incident. They should not be added to unrelated national observations, generalized across all 13 countries, or presented as the percentage of the entire West African Internet that disappeared.

The following day's Ghanaian notice stated that all four landing providers were cut off from international data services. [3] On its face, that language may look different from the earlier estimates, which left some residual percentage on three systems. The records need not be inconsistent. Cable-system capacity, usable capacity delivered at a particular landing provider, and end-to-end international service are different operational layers. They were also described at different moments.

Residual system capacity may be inaccessible to a provider, insufficient for service, isolated by another fault, or overtaken by a changing incident state.

The same discipline applies to recovery. Cloudflare's later quarterly analysis described uneven traffic restoration and relayed reports of shifts toward Equiano and Maroc Telecom West Africa capacity. [12] Operator announcements described rerouting and additional circuits. [13][15] Ghana's regulator described capacity obtained through unaffected regional partners and content available through local caches. [3] Each is evidence of a restoration mechanism, but none establishes a uniform moment when all customers, applications, or countries returned to normal.

Service restoration also cannot be equated with cable repair. An affected network might regain partial international access through emergency transit while a damaged system remains unrepaired for weeks. A website might become reachable when its prefix is announced again, even as congested links continue to impair other services. A bank's domestic transaction flow might improve because a dependency is locally hosted, while an overseas authentication or fraud-control service remains slow.

The relevant clocks include physical fault observation, route withdrawal, alternate-circuit activation, traffic recovery, application recovery, vessel dispatch, splice completion, and return to regular cable operation.

A publication-grade incident record should therefore resist the appeal of one headline number. The reviewed evidence supplies a Ghanaian capacity estimate, landing-provider service statements, country-level traffic observations, operator restoration reports, and cable-specific repair dates. It does not supply a complete regional loss total, a customer-by-customer outage matrix, or a single percentage that accurately describes every layer. Preserving those limits makes the continuity analysis stronger because it shows exactly which controls can be evaluated and which remain unproven.

Control question one: Were the physical routes genuinely independent?

Ghana's final repair update reported a finding attributed to the subsea cable service providers: all four faults occurred where the systems crossed the Le Trou Sans Fond submarine canyon off Abidjan. [7] Internet Society's analysis treated that convergence as a physical single point of failure for the regional event. [9] This is the clearest public basis for concluding that separately named systems shared a consequential corridor risk. It is not, however, a complete survey of their alignments or a forensic explanation of the damage mechanism.

Physical diversity is often described at continental scale. Two cables may connect different landing countries, have different owners, or follow visibly different ocean routes for most of their length. Those differences are valuable, but continuity depends on their closest common exposure. If several systems converge at a canyon crossing, landing approach, terrestrial duct, power dependency, or maintenance bottleneck, the shared segment may dominate their combined risk. The March event shows why route-diversity analysis must examine common hazards at the scale at which damage can occur.

The public record leaves important geometric questions unanswered. It does not disclose complete cable coordinates, precise separation distances, burial conditions, armor specifications, maintenance histories, or detailed seabed surveys for the four fault locations. It does not show whether the crossings were clustered within one narrowly bounded area or distributed across multiple points exposed to a related hazard. Nor does it publish a cable-by-cable forensic report establishing when and how each system was damaged.

Those gaps matter because a common corridor does not automatically prove a single physical trigger. Four faults at canyon crossings could be associated with one geophysical event, multiple related movements, independent damage within a hazardous area, or another mechanism not established in the reviewed sources. Early public descriptions also differed about fault locations. The final account supports a shared-exposure conclusion, but it does not justify a claim of sabotage, intent, negligent route design, or one precisely identified seabed cause.

The appropriate accountability question is evidentiary. Cable-system operators should be able to show bounded route maps to authorized assessors, identify common hazard zones, document route-separation assumptions, and explain how those assumptions informed restoration planning. Where public disclosure would create security or commercial risks, regulators or independent technical assessors can examine restricted evidence and publish conclusions at a safer level of detail. Accountability does not require exposing exact cable coordinates to everyone; it requires a credible method for testing diversity claims.

The same analysis must extend beyond wet plant. A cable may follow an independent marine alignment yet converge with another system at a landing station, backhaul route, interconnection facility, or cross-border terrestrial corridor. Conversely, two systems with some marine proximity may feed different carriers and restoration paths that reduce service impact. A rigorous assessment maps the complete service route from cable segment through landing infrastructure, terrestrial transport, transit provider, exchange point, and customer network.

The March outage demonstrates that counting systems without examining their dependency graph can materially overstate resilience.

Route design also interacts with repair design. Systems exposed to a common hazard may depend on the same regional vessel availability, permitting environment, maintenance agreement, depot, or specialist workforce. The 2024 response eventually used separate vessel movements and joint commissioning among several systems, but the need for coordination itself reveals another concentration dimension. An independent route is more valuable when its monitoring and repair dependencies are not all subject to the same delay.

The incident does not establish that every shared dependency was avoidable. Submarine cable routes face seabed geography, landing access, environmental constraints, cost, permitting, and engineering tradeoffs. The accountability standard is not a retrospective demand for zero shared risk. It is whether known common-mode exposure was identified, bounded, incorporated into continuity planning, and counterbalanced by alternate capacity that could actually carry traffic when the primary corridor failed.

Control question two: Could networks reach and use unaffected transit?

Once international capacity disappeared, physical alternatives mattered only through operating networks. Border Gateway Protocol is the principal mechanism through which autonomous systems exchange reachability information across administrative boundaries. An autonomous system announces the IP prefixes it can reach, learns routes from neighbors, and applies policy when selecting and propagating paths. RFC 4271 defines this inter-domain reachability framework. [17] It does not prescribe the commercial relationship behind a route or prove what policy any West African operator used during the outage.

This distinction explains why the presence of an unaffected cable does not guarantee restoration. A carrier must have access to the cable through its own facilities or a supplier. The required circuit must exist, or be capable of rapid activation. The carrier and its neighbors must exchange usable routes. Filters, preferences, communities, maximum-prefix controls, and other operational safeguards must permit a controlled change without accepting unsafe reachability. The alternate link must then have enough capacity to carry diverted demand at an acceptable level of congestion.

Operational BGP guidance such as RFC 7454 addresses practices including filtering, route handling, and resilience at interconnection boundaries. [18] It provides a framework for understanding the controls involved when networks shift traffic, but it is not evidence that any named operator complied with or violated a particular practice. The incident sources do not publish complete router configurations, route-policy changes, session states, or update streams for every affected autonomous system.

Cloudflare did observe a revealing slice of this layer. Part of the IPv4 address space originated by MainOne's AS37282 stopped being announced between approximately 07:30 and 15:00 UTC. Cloudflare reported that reachability to a MainOne-hosted nameserver and website disappeared and later returned in the same broad period. [11] That observation connects physical infrastructure failure to the control plane: a service can be unavailable not only because packets encounter congestion, but because the Internet temporarily lacks a route to the relevant address space.

The observation must remain bounded. It does not identify every prefix affected, every upstream session, the cause of each withdrawal, or the internal decision that restored announcements. It does not reveal whether alternate paths were rejected, absent, deliberately suppressed, or simply unable to carry the affected service. Nor does it reconstruct the route history of every MainOne customer. It is evidence that BGP reachability changed during the disruption, not a complete attribution of the change.

Other sources show alternatives functioning in practice. Internet Society reported that Moov Africa Ivory Coast maintained access through the Maroc Telecom West Africa cable, whose route differed from those of the four damaged systems. [10] That case is important because it demonstrates more than installed infrastructure. It shows a network retaining access through a route that remained physically available and operationally usable. The contrast with more severely affected networks supports a route-diversity conclusion without proving that every provider had the same opportunity or capacity.

Cloudflare's quarterly summary reported traffic shifts toward Equiano and Maroc Telecom West Africa capacity as networks responded. [12] Those reported shifts indicate that unaffected systems absorbed traffic, but public data do not fully separate pre-contracted capacity from emergency purchases or quantify spare headroom across each route. A reported increase on one system should not be treated as an independently audited regional capacity measurement. What can be concluded is that alternate submarine routes carried meaningful diverted demand.

Angola Cables provided another concrete route example. It said it used unaffected SACS capacity to redirect international traffic across the South Atlantic to Brazil and onward toward the United States and Europe. [15] That detour may be longer than a direct northbound route, but its value lay in belonging to a different physical and interconnection chain. At the time of its statement, the operator also said the cause of the West African cable disruption had not been confirmed, reinforcing the need to separate restoration action from physical-cause claims.

Bayobab, part of MTN Group, said it rerouted traffic and enabled additional circuits while coordinating repair activity after faults affected the four systems. [13] This supports the existence of carrier-level mitigation, but it does not disclose route-level logs, customer restoration times, reserved capacity, or congestion on the replacement circuits. An operator statement can establish an action while leaving the quality and distribution of the resulting service unknown.

Cross-border terrestrial paths were especially important for countries without direct access to every landing system. Internet Society's analysis describes terrestrial fallback and the dependence of inland markets on routes through coastal neighbors. [9] Such connectivity can diversify cable access, but it can also create new common dependencies. Two upstream services sold under different names may converge on the same coastal landing, terrestrial fiber, border crossing, or transit network. Commercial supplier diversity is therefore not equivalent to end-to-end route diversity.

Capacity is the final constraint. A technically reachable alternate route may preserve messaging and basic browsing while failing under the full traffic load that previously crossed several high-capacity systems. Congestion can cause latency, packet loss, timeouts, and application failure even when BGP continues to advertise reachability. The NCA's later coordination with banks and utilities occurred precisely because constrained bandwidth had consequences for services that needed more than the existence of a route. [5]

The operating test for transit diversity consequently has several parts: a physically surviving route, a commercial right to use it, a working circuit, acceptable routing policy, sufficient capacity, and evidence that traffic actually moved. The 2024 record supplies examples where that chain worked and places where service remained constrained. It does not support a universal judgment about any operator because the contractual, routing, and congestion evidence needed for that judgment is not public.

Control question three: What could local exchange and caching preserve?

Ghana's response illustrates both the value and the limits of locality. The NCA said operators obtained capacity from unaffected regional partners and relied on local caches, including content available at the Ghana Internet Exchange. [3] Local content could remain reachable without traversing the damaged international systems if the user network, content network, and exchange fabric retained connectivity. That reduced demand on scarce international links and preserved access to eligible services.

An Internet exchange point provides a place where participating networks can exchange traffic through bilateral sessions or shared route-server arrangements. RFC 7947 explains common BGP operations at exchanges, including the distinction between bilateral peering and route-server service. [19] It does not establish the configuration of GIX, AIX, or any participating network in March 2024. Public sources do not reveal every session, announced prefix, cache fill state, or participant policy at those exchanges.

The boundary is essential. An IXP does not manufacture an international path. If a requested service, authentication endpoint, payment processor, software dependency, or authoritative system is hosted abroad, local exchange alone cannot make it available. A cache can serve objects already present within its operating rules, but it cannot substitute for every dynamic application. Local peering reduces unnecessary international carriage; it does not eliminate international dependency.

The NCA's continuity update shows why those distinctions mattered outside the telecommunications sector. It convened banks, payment and securities institutions, water and electricity utilities, mobile operators, cable providers, GIX, and the Accra Internet Exchange because constrained bandwidth was affecting banking transactions and digital utility payments. Participants considered prioritizing limited data capacity for critical services and recommended more local hosting and exchange participation.

[5] These were documented continuity responses, not proof that every institution used an exchange or that local hosting would have prevented every failure.

Locality is nevertheless a material control. If a public service hosts its domestic portal, identity dependency, payment component, and authoritative DNS through arrangements reachable inside the country, it may reduce the number of international links required for a transaction. If only the front page is local while core APIs and authentication remain overseas, the service may appear present but fail when used. Effective continuity testing must therefore follow complete application dependencies rather than count locally hosted pages.

Exchange participation also changes the allocation of scarce capacity. Traffic that stays local frees international links for services that genuinely require them. During a multi-cable disruption, that can increase the practical value of every remaining gigabit. The benefit depends on participation, route acceptance, content presence, port and fabric capacity, and the health of domestic backhaul. It should be measured through traffic and service evidence rather than assumed from the existence of an exchange.

For accountability purposes, IXPs and content networks can document fabric availability, route-server health, aggregate traffic changes, cache availability, and incident communication within appropriate confidentiality limits. Participating networks can retain session-state and route-change records. Critical institutions can map which dependencies stayed local and which crossed international transit. Those records would show whether locality worked as a continuity layer without implying that the exchange was responsible for failures beyond its operational boundary.

The March event therefore supports a precise conclusion: GIX, AIX, caches, and local hosting could preserve or accelerate some domestic access, and their use reduced pressure on damaged international connectivity. They could not replace the missing capacity of four major systems. Treating locality as one layer in a broader resilience design is consistent with the evidence; presenting it as a complete answer would not be.

Control question four: How were restoration and physical repair governed?

The wet-plant response began with uncertainty. In its early sequence of updates, Ghana's regulator relayed preliminary fault locations that changed as operators refined their information. By 16 March, according to the NCA's account, operators had remotely approximated fault locations, were preparing repair vessels, and initially expected restoration to require at least five weeks after vessel dispatch. [4] Remote localization was a planning input, not a completed seabed inspection.

Temporary connectivity work proceeded in parallel. Operators sought unaffected capacity, rerouted traffic, activated circuits, and relied on local exchange and caches. [3][13][15] These measures could improve customer reachability while the cables remained physically severed. Their success should be evaluated through traffic, route, congestion, and application evidence, not inferred from a later repair announcement.

The NCA's 28 March update described two repair arrangements. SAT-3 was to use a vessel sailing from Cape Town. ACE, WACS, and MainOne jointly commissioned another vessel sailing from London. The regulator published tentative cable-by-cable schedules while emphasizing the practical dependencies involved in vessel mobilization and repair. [6] Those dates were forecasts, not completion evidence.

Joint commissioning can reduce duplicated mobilization and coordinate work in a shared region, but it also requires priority decisions among systems. A useful accountability record would show when each operator confirmed the fault, requested maintenance service, secured permits and spares, mobilized a vessel, reached the work site, completed inspection and splicing, tested the cable, and returned it to normal operation. The public sources provide only part of that sequence.

The NCA's final update supplies the strongest completion record. It reported SAT-3 repaired on 6 April, ACE on 17 April, WACS on 29 April, and MainOne on 8 May. [7] These dates must remain cable-specific. They show a repair campaign extending over more than a month after the initial regional disruption and do not identify one common restoration date for every service.

MainOne separately confirmed completion of its repair and a return to regular operation. It also said it had supplied restoration capacity and rerouted traffic during the outage. [14] The two parts of that statement describe different clocks: interim continuity through other capacity, followed by restoration of the damaged system. MainOne's announcement does not provide customer-by-customer service records or independently establish when every dependent network recovered.

The final regulator account also recorded post-incident redundancy directives. [7] Their existence shows that continuity planning remained an active regulatory concern after physical repair. Public evidence does not establish whether every directive has since been implemented, funded, exercised, and independently tested. A policy announcement should not be treated as durable remediation until technical evidence demonstrates changed dependencies and successful failure exercises.

Repair governance is consequently more than vessel speed. It includes monitoring quality, fault localization, maintenance agreements, vessel access, spares, permitting, multi-owner coordination, communication, and post-repair testing. It also includes the restoration arrangements that carry traffic during the wet-plant interval. A resilient system needs both: rapid access to alternate capacity for immediate continuity and credible physical repair capability for sustained recovery.

Control question five: What evidence should each controller retain?

The event divided practical control among several actors. No single participant controlled seabed routes, international circuits, BGP policy, exchange participation, local application hosting, regulatory coordination, and vessel repair. Accountability should follow those boundaries. It should ask what each controller knew, what it could change, which continuity claim it made, and what contemporaneous evidence can verify performance.

Controller Practical control Evidence that should exist Boundary of inference
Cable-system operators Route engineering, system monitoring, maintenance arrangements, fault localization, repair commissioning, and return-to-service testing Bounded route and hazard maps, alarm chronology, localization results, maintenance requests, vessel milestones, splice and optical-test results, and common-mode remediation tests Possession of these controls does not prove that a fault was avoidable or negligently handled
Mobile, fixed, and wholesale networks Upstream selection, alternate circuits, BGP policy, traffic engineering, congestion management, and customer communication Capacity contracts, circuit inventories, activation records, BGP changes, flow and utilization data, restoration matrices, and timestamped notices A surviving route does not prove adequate capacity or equal recovery for every customer
IXPs and content networks Exchange-fabric availability, route-server or bilateral interconnection support, cache presence, and participant communication Fabric health, aggregate traffic changes, session availability, cache status, and incident communications An exchange cannot restore traffic whose required destination or dependency remains international
Regulators Continuity requirements, cross-sector coordination, information demands, and remediation oversight Incident notices, evidence requests, service-priority decisions, directive tracking, exercises, and independent verification A directive does not prove implementation or create a retrospective legal finding
Banks, utilities, and public institutions Hosting architecture, application dependencies, carrier diversity, and continuity priorities Dependency maps, carrier and hosting arrangements, application tests, transaction impacts, and failover results Local hosting can reduce exposure but cannot remove every international dependency

For cable-system operators, the core question is whether route diversity was analyzed at the scale of shared hazards. A useful record would identify where systems converge, what physical and operational assumptions support acceptable separation, and which alternatives remain if the common corridor is lost. Exact coordinates need not be broadly published. A regulator or qualified assessor can verify sensitive evidence while releasing a bounded conclusion about whether common-mode risk was identified and mitigated.

Operators should also retain a cable-specific event chronology. Alarm timestamps, remote test results, fault-location estimates, maintenance-authority communications, vessel requests, permit dependencies, arrival times, repair stages, and optical acceptance tests would distinguish unavoidable physical delay from organizational delay. The reviewed sources confirm coordination and final dates but do not expose this complete chain. Its absence from public reporting is an evidence gap, not proof of misconduct.

For carriers and access networks, continuity evidence begins before the incident. Alternate-capacity contracts should identify provider, physical route where known, committed capacity, activation conditions, lead time, and whether two suppliers ultimately depend on the same cable or landing infrastructure. During an event, circuit activation logs and utilization measurements should show what capacity became available, when it carried traffic, and where congestion persisted.

BGP and traffic records are equally important. Networks should retain relevant route announcements and withdrawals, session changes, policy changes, traffic shifts, and rollback decisions. These records can demonstrate whether an alternate circuit was technically reachable and accepted without implying that every policy change must be public. Aggregated or independently assessed evidence may protect security and commercial information while still verifying the continuity outcome.

A restoration matrix should distinguish customers, services, and layers. It could record when basic IP reachability returned, when capacity crossed an agreed threshold, when critical applications became usable, and when regular routing resumed. This avoids declaring an entire network restored because one prefix reappeared or one traffic graph rose. It also exposes whether scarce capacity was distributed according to documented priorities.

For IXPs and content networks, evidence should focus on their actual boundary. Fabric uptime, route-server health, bilateral session availability, traffic volume, cache-service status, and participant communications can show what remained local. These records should not be used to assign an exchange responsibility for overseas content or a participant's absent route. They should show how much avoidable international traffic was removed and which local services remained reachable.

Regulators can join these evidence domains without pretending to operate the networks. Ghana's updates documented coordination, capacity constraints, critical-service concerns, vessel arrangements, completion dates, and redundancy measures. [2][4][5][6][7] A stronger post-incident record would connect each directive to a responsible owner, completion date, technical test, and independent result. It would also preserve a clear distinction between confidential evidence examined by the regulator and conclusions supported by material released publicly.

Critical institutions have their own control surface. A bank or utility may buy two carrier services but still depend on the same international route, overseas cloud region, identity provider, DNS service, or payment interface. Dependency mapping and realistic exercises can reveal that convergence. The NCA's cross-sector meeting demonstrates that cable resilience is also an application-continuity issue, but the public record does not disclose the architecture or recovery of each institution. [5]

Evidence quality matters because many continuity claims are easy to state after an event. A route map can show nominal diversity; a contract can show theoretical capacity; a BGP table can show reachability; an exchange graph can show local traffic; and a repair notice can show physical completion. None alone proves end-to-end service continuity. The strongest account links them by time: a fault occurred, a route changed, capacity moved, congestion evolved, services recovered, a vessel repaired the system, and remediation was later exercised.

Distributed control is not distributed blame

The incident created multiple accountability questions without establishing wrongdoing. Cable owners controlled aspects of route design, monitoring, maintenance arrangements, fault localization, and repair. Carriers and access networks controlled supplier selection, alternate circuits, routing policy, traffic engineering, and communication with customers. Exchanges and content networks controlled local interconnection and cache availability within their participant boundaries. Regulators controlled coordination and evidence requirements. Critical institutions controlled some carrier, hosting, and application-dependency decisions.

Those roles identify where evidence should be found. They do not establish negligence, sabotage, malicious intent, or an event-specific legal breach. Public sources do not show the complete information available to each actor before the incident, the contractual duties between them, the engineering constraints on route selection, or the reasonableness of each response under applicable law. A technical accountability analysis must not convert control into fault without that evidence.

The same caution applies to national comparisons. A country with more visible traffic loss may have had fewer independent cable routes, less spare capacity, different traffic composition, or greater dependence on one provider. A country that recovered faster may have had pre-contracted capacity, favorable interconnection, or lower demand on an alternate path. Public traffic graphs do not isolate all those variables. They are evidence of outcome, not a complete causal score.

Moov Africa Ivory Coast's continued access through Maroc Telecom West Africa is a particularly useful counterexample because it demonstrates that different route geometry and usable connectivity mattered. [10] It does not prove that every other provider could have purchased the same service on the same terms or activated it in time. Angola Cables' SACS detour likewise demonstrates the value of a distinct international direction without proving that it was suitable for every affected flow. [15]

Accountability should therefore be proportionate to control and evidence. A cable operator can be asked to demonstrate common-mode analysis. A carrier can be asked to demonstrate alternate-capacity readiness and routing behavior. An exchange can be asked to demonstrate local fabric availability. A regulator can be asked to demonstrate follow-through on resilience directives. A bank or utility can be asked to demonstrate application continuity. Each question is meaningful without presuming the answer.

This approach also avoids turning recommendations into retrospective duties. More route diversity, local hosting, exchange participation, and tested capacity may be prudent responses to the event. Whether any particular actor was legally required to implement them before March 2024 is a separate question requiring applicable law, contracts, standards, and facts not established here. The public record supports a control analysis and a demand for verifiable remediation, not a legal verdict.

The unknowns that limit stronger conclusions

Several material questions remain unresolved in the reviewed public record:

  • The exact physical fault time and sequence for ACE, MainOne, SAT-3, and WACS are not consistently disclosed.
  • Complete cable coordinates, separation distances, burial conditions, maintenance histories, and forensic seabed findings are not public.
  • Provider-by-provider BGP updates, route-selection decisions, circuit activations, and congestion measurements remain unavailable.
  • The amount of pre-contracted, reserved, or emergency-purchased restoration capacity on each alternate route is unknown.
  • Customer-level downtime, application recovery, and complete economic losses across the 13 affected countries have not been established.
  • The evidence does not establish negligence, sabotage, malicious intent, or an event-specific legal violation by an operator.
  • Public material does not prove that every post-incident redundancy measure has been implemented and independently exercised.

These gaps prevent several tempting but unsupported conclusions. The canyon-crossing finding cannot be expanded into an exact geophysical mechanism. A traffic recovery graph cannot establish complete service restoration. An operator's rerouting statement cannot establish adequate capacity for every customer. A prefix withdrawal cannot reveal an entire network's BGP policy. A regulator's directive cannot establish durable remediation.

The gaps also define a practical disclosure agenda. Authorities do not need to publish sensitive topology or contracts in full. They can report whether alternate routes are physically independent at material hazard points, whether committed restoration capacity exists, how quickly it can be activated, whether realistic exercises have succeeded, and whether critical services can function under constrained international connectivity. Independent verification can provide confidence without exposing operational secrets.

A complete regional economic loss total would require consistent service, customer, and sector data across all affected countries. The sources do not provide that dataset. Banking and utility-payment effects in Ghana are documented, as are substantial country-level traffic disruptions, but extrapolating them into one monetary total would exceed the evidence. [5][8][11][16] The absence of a total should not obscure the seriousness of the incident; it should prevent false precision.

Uncertainty about cause likewise does not erase the resilience lesson. Even without a complete seabed forensic report, four important systems became unavailable or impaired near a shared corridor, and networks with usable alternatives experienced different continuity outcomes. That is enough to test route-diversity claims. It is not enough to assign blame for the physical damage.

Why this case is different from nearby cable and routing incidents

The 2024 West African event should not be collapsed into other well-known outage patterns. Tonga's 2022 crisis involved a country dependent on a sole international cable, volcanic damage, and national-scale repair logistics. The central question there was how a single international route and difficult physical repair shaped national isolation. West Africa in 2024 involved several named transnational systems and a regional common corridor, with continuity varying according to alternate cable, transit, and terrestrial options.

Shetland's 2022 disruption involved two island routes impaired within a short period and service outcomes that depended on provider-specific access to surviving capacity. That case is useful for understanding access to remaining routes, but its island topology and bounded provider environment differ from a 13-country region linked through multiple coastal and inland networks.

The 2018 Google and MainOne incident concerned a BGP route leak and upstream acceptance. Its primary failure plane was routing information rather than simultaneous or near-simultaneous physical impairment of four submarine systems. BGP remains important in the 2024 analysis because alternate reachability had to operate, but the initiating capacity shock was physical.

The bounded subject here is therefore distinctive: multiple transnational systems shared a consequential West African corridor, and the resulting service outcomes exposed the relationship among physical separation, contracted transit, BGP reachability, circuit capacity, cross-border terrestrial paths, exchange participation, local caching, and coordinated repair. Removing those network controls would leave only a generic cable-damage story and would erase the evidence that explains why continuity differed.

The operational test for regional route diversity

The evidence supports a six-part test. First, alternatives must be physically independent enough that one plausible hazard does not remove all of them. This requires analysis of marine corridors, landing approaches, terrestrial backhaul, facilities, power, and repair dependencies. Independence need not be absolute, but shared exposure must be known and bounded.

Second, alternative capacity must be commercially available. A cable shown on a map is not a continuity resource for a network that lacks a contract, cross-connect, supplier arrangement, or emergency activation mechanism. Evidence should distinguish pre-provisioned capacity from capacity acquired after the incident because the two imply different recovery times and confidence.

Third, the route must be technically reachable. BGP sessions, prefix announcements, filters, preferences, and operational safeguards must allow traffic to move without creating a new routing incident. The AS37282 observation shows why this layer matters, even though it does not reveal the complete route history. [11][17][18]

Fourth, the alternative must have sufficient capacity under realistic demand. Mere reachability can coexist with severe congestion. Operators should test failure loads, identify priority services, and measure packet loss, latency, utilization, and application performance. Ghana's constrained-bandwidth response demonstrates the difference between some connectivity and enough connectivity for critical transactions. [5]

Fifth, interconnection and locality must work within their actual boundaries. Cross-border terrestrial paths can connect inland networks to unaffected landings. IXPs and caches can preserve eligible local traffic and release international capacity for remote dependencies. Neither should be credited until route, traffic, and service evidence shows that it performed during the failure. [3][9][19]

Sixth, the design must be exercised. A continuity plan becomes credible when circuits activate within a measured time, routes converge safely, capacity carries realistic traffic, critical applications work, and responsible teams retain evidence. Post-incident remediation is durable only when a later test demonstrates that the dependency graph has changed or that the surviving alternatives can absorb the expected load.

West Africa's March 2024 outage met the threshold for a regional accountability test because it exposed every link in this chain. The four cable names did not guarantee four independent continuity paths. Some networks reached unaffected systems, some traffic shifted through distinct international directions, local exchange preserved part of the reachable service set, and physical repairs proceeded on separate cable-specific schedules. [7][10][12][13][14][15]

The final lesson is narrow but consequential. Regional route diversity exists only when alternatives survive the same event, can be bought or activated, remain reachable through routing policy, provide enough capacity, connect through functioning cross-border and exchange arrangements, and succeed in realistic operation. The public evidence shows that those controls materially shaped the 2024 outcome. It does not establish negligence or prove that remediation is complete. The next accountability step is not another cable count; it is verifiable evidence that the region's alternate routes now work as independent continuity systems.

Sources

  1. [1] Ghana National Communications Authority — Undersea Cable Disruptions Affect Data Services
  2. [2] Ghana National Communications Authority — Update 2
  3. [3] Ghana National Communications Authority — Update 3
  4. [4] Ghana National Communications Authority — Update 4
  5. [5] Ghana National Communications Authority — Update 5
  6. [6] Ghana National Communications Authority — Update 6
  7. [7] Ghana National Communications Authority — Update 7
  8. [8] Internet Society Pulse — Major Internet Outages Across Western and Southern Africa
  9. [9] Internet Society — 2024 West Africa Submarine Cable Outage Report
  10. [10] Internet Society Pulse — Morocco's Role in Regional Connectivity
  11. [11] Cloudflare — Undersea Cable Failures Cause Internet Disruptions Across Africa
  12. [12] Cloudflare — Q1 2024 Internet Disruption Summary
  13. [13] MTN Group and Bayobab — Connectivity Disruptions Due to Undersea Cable Damage
  14. [14] MainOne — Successful Repair of Submarine Cable
  15. [15] Angola Cables — Disruption to International Submarine Cables off the West African Coast
  16. [16] Associated Press — Internet Outages Reported Across Africa
  17. [17] RFC 4271 — A Border Gateway Protocol 4
  18. [18] RFC 7454 — BGP Operations and Security
  19. [19] RFC 7947 — Internet Exchange BGP Route Server