The move from carrier MPLS to SD-WAN over the internet is usually justified on cost and agility, but it quietly swaps one trust model for another, and many organisations do not notice which risks they have given up and which they have taken on. MPLS asks you to trust a carrier's core network and its staff; SD-WAN asks you to trust your own encryption, your device admission and, above all, a small set of controllers and orchestrators that can reconfigure every site at once. This guide compares the two on encryption assumptions, attack surface, management plane identity and migration risk, and gives decision criteria and a worked scenario. It is for network and security architects deciding whether, and how, to replace or augment a carrier WAN. Hardening detail for each option lives in the SD-WAN Security Guide and the MPLS and Carrier WAN Security Guide; this page is about the choice between them.
Key takeaways
- MPLS separates traffic; it does not encrypt it. RFC 4381 states that BGP/MPLS IP VPNs themselves provide no encryption, integrity or authentication, and that CE-to-CE IPsec is the only way it identifies to protect a VPN fully against insiders.
- SD-WAN encrypts by design, but moves the crown jewels. The overlay protects packets on the underlay, while the orchestrator and controllers become the place where one compromise reaches every branch.
- "Private" versus "internet" is the wrong question. The real question is who can read, redirect or reconfigure your traffic, and both models answer it with a trust assumption.
- Management plane identity decides the SD-WAN risk. Admin accounts, API tokens, controller peer identities and device certificates are what attackers abused in exploited SD-WAN controllers.
- Migration is the riskiest phase. Running both networks side by side doubles the paths, and local internet breakout changes where inspection happens on day one.
The short answer
Neither is inherently "more secure". MPLS gives route separation inside a carrier network you cannot inspect, with no cryptographic protection unless you add it. SD-WAN gives encryption you control over transports you do not trust, with a concentrated control plane that you must defend like an identity system. If you stay on MPLS for sensitive traffic, add CE-to-CE IPsec or plan to rely on application-layer encryption. If you move to SD-WAN, budget as much effort for orchestrator hardening, admin identity and peer monitoring as for the transport change, and decide before cutover where internet-bound traffic will be inspected. Many estates end up hybrid: SD-WAN overlay across both MPLS and internet underlays, which keeps carrier SLAs where needed and adds encryption everywhere.
Encryption assumptions
The most common misconception in WAN security is that a carrier VPN is private in the cryptographic sense. RFC 4364, the Standards Track specification for BGP/MPLS IP VPNs (February 2006), says the methods it describes do not by themselves encrypt data for privacy. RFC 4381, the security analysis of the same architecture, lists what the service gives (address and routing separation through per-customer VRFs, hiding of the provider core, resistance to label spoofing from customer edges) and what it does not (confidentiality, integrity, origin authentication, protection against core misconfiguration or insiders). If the provider is not trusted, RFC 4381 recommends IPsec between the customer's own edge devices.
SD-WAN inverts the starting point. The overlay is built from encrypted tunnels between your edges, so the underlay (broadband, LTE or 5G, MPLS) is treated as untrusted. That matches the posture in the UK NCSC's IPsec guidance, which applies to data crossing the internet and commercial WAN circuits alike and makes no assumptions about the security of the bearer network. Its recommended profile is IKEv2 with AES-GCM, ECP group 19 or MODP group 14, certificate authentication and defined key lifetimes; IKEv1 is strongly advised against. Ask any SD-WAN supplier how its tunnels map to a profile of that kind, because "encrypted" on a datasheet says little about key exchange, authentication or algorithm negotiation.
| Question | MPLS L3VPN | SD-WAN over internet |
|---|---|---|
| Is traffic encrypted by default? | No; separation only | Yes, between your edges, if the overlay is configured as designed |
| Who could read traffic in transit? | Anyone with access to carrier core links or devices, including insiders | Nobody on the underlay without the keys; anyone who controls the key distribution or a trusted edge |
| Who could redirect traffic? | The carrier, through VRF or route target changes or a compromised provider edge | Whoever controls the orchestrator or controllers (policy and route push) |
| Integrity and origin authentication | Not provided by the service | Provided by the tunnel protocol, if peers are authenticated with strong identities |
| Where keys live | Nowhere, unless you add CE-to-CE IPsec | On edges and, in many designs, in controllers that distribute them |
| Inter-provider exposure | Every carrier in the path must be trusted, per RFC 4381 | Underlay providers carry only ciphertext |
Attack surface compared
Moving to SD-WAN does not reduce attack surface so much as relocate it. With MPLS, the most powerful positions (provider edge and core routers, carrier management systems) belong to someone else and are hidden from you. With SD-WAN, the most powerful positions are yours, and some are reachable from the internet.
| Surface | MPLS | SD-WAN | Relevant ATT&CK technique |
|---|---|---|---|
| Transport path | Carrier core; taps or insiders could read cleartext | Public internet carrying ciphertext; exposed to volumetric attacks on edge addresses | T1040 Network Sniffing |
| Routing control | Carrier BGP and VRF configuration | Your controllers distributing routes and policy | T1557 Adversary-in-the-Middle |
| Central management | Carrier systems and your carrier portal accounts | Your orchestrator or manager, its GUI, API and peering services | T1601 Modify System Image (software downgrade) |
| Branch device | Customer edge router, often carrier-managed | SD-WAN edge with direct internet exposure and local breakout | T1599 Network Boundary Bridging |
| Supplier | Carrier staff, subcontracted access circuits, inter-provider links | SD-WAN vendor cloud control plane (if cloud-delivered) and any managed service provider | Supplier risk, not a single technique |
Carrier compromise is not hypothetical. Cisco Talos reported in February 2025 that the group it tracks as Salt Typhoon accessed core infrastructure at several US telecommunications providers, largely with stolen legitimate credentials, captured authentication traffic between devices, and in one case kept access for over three years. See the Salt Typhoon telecom intrusions breach analysis. For an MPLS customer, that is exactly the insider-equivalent position RFC 4381 warns about.
SD-WAN controllers are targets too. Orchestrators and controllers from more than one supplier appear in the CISA Known Exploited Vulnerabilities catalogue, including Cisco Catalyst SD-WAN Controller and Manager (CVE-2026-20127, added February 2026) and Versa Director (CVE-2024-39717, added August 2024); suppliers are listed alphabetically. Cisco Talos reported that the actor exploiting CVE-2026-20127, which it tracks as UAT-8616, had been active since at least 2023 and left indicators such as rogue control-connection peers, unauthorised SSH keys and cleared logs. CISA issued Emergency Directive 26-03 for US federal agencies running affected systems. The lesson for the comparison is structural: whichever supplier you choose, the controller tier concentrates fabric-wide control in a few systems.
Management plane identity
In an MPLS service the identities that can change your WAN are mostly the carrier's: its engineers and its provisioning systems. Your exposure is the carrier portal, the change request process and whoever can raise a ticket in your name. In SD-WAN those identities become yours to govern, and most of them are non-human.
| Identity | MPLS equivalent | SD-WAN control to apply |
|---|---|---|
| Human administrators of the fabric | Carrier engineers (outside your control); your portal users | Phishing-resistant MFA, role separation between policy, security and monitoring, named accounts only |
| API tokens and automation accounts | Carrier provisioning systems | Scoped tokens, inventory with owners, rotation, alerting on use from new sources |
| Controller-to-controller and internal service accounts | Not visible to you | Monitor logins and key changes; treat any interactive use as an incident signal |
| Edge device identity (certificates, serial allowlists) | Physical circuit handoff | Certificate lifecycle management, allowlist reconciliation, prompt revocation of lost or returned devices |
| Control-connection peers | Carrier BGP peers | Reconcile every new peer against inventory and change records |
The Talos recommendations for CVE-2026-20127 read as an identity checklist: verify every control-connection peering event against maintenance windows and known addresses, audit SSH authorised keys for privileged accounts, and look for unexplained root sessions and account changes. See the Machine Identity, PKI and Certificate Lifecycle Guide and Network Device Management Plane Hardening Guide.
Threat model, failure modes and controls by option
MPLS
- Threat model: confidentiality and integrity depend on the carrier's core, staff and suppliers; availability depends on carrier SLAs, which are usually strong.
- Architecture in words: your customer edge router peers with a provider edge router (statically or with BGP); the carrier places your routes in a VRF and carries labelled packets across its core to your other sites.
- Failure modes: route target misconfiguration leaking routes between customers, compromised provider edge or core devices, unencrypted sensitive traffic, and carrier portal accounts without MFA.
- Controls: CE-to-CE IPsec for sensitive flows, authenticated CE-PE routing with prefix limits, carrier assurance clauses, and independent monitoring of reachability and routes.
SD-WAN over internet
- Threat model: the underlay is hostile but carries only ciphertext; confidentiality, integrity and routing depend on your controllers, key distribution and device admission; availability depends on internet path diversity.
- Architecture in words: edges at each site build encrypted tunnels over any transport; controllers authenticate edges, distribute routes, policy and often keys; an orchestrator pushes configuration and software; local breakout sends some traffic straight to the internet.
- Failure modes: orchestrator or controller exposed to the internet, a rogue peer admitted through weak device identity, shared or stale admin accounts, local breakout without inspection, and logs stored only on the devices an attacker can clear.
- Controls: management plane restricted to trusted networks, MFA for every administrator, certificate-based edge identity, off-box logging, peer reconciliation, and a defined inspection point (on-box, regional hub or cloud security service) for breakout traffic.
Internet breakout changes where you inspect
With MPLS, internet traffic typically backhauls to a data centre or regional hub where firewalls, proxies and data loss prevention sit. SD-WAN makes local breakout easy, which improves performance for SaaS and cloud traffic and removes that inspection point unless you replace it. Decide per traffic class before cutover:
- Trusted SaaS and cloud destinations: local breakout with security service steering or on-box filtering, as policy allows.
- General web browsing: breakout only through a cloud security service or an inspecting edge, never direct.
- Internal applications and sensitive data flows: keep inside the overlay to data centre or cloud.
- Guest and IoT segments: separate segments with their own breakout policy, never sharing the corporate overlay.
For the security service layer on top of SD-WAN, see SASE vs SD-WAN and the SASE and SSE Guide.
Decision criteria
| Factor | Favours staying on MPLS (with IPsec) | Favours SD-WAN |
|---|---|---|
| Latency and jitter needs | Real-time or industrial traffic needing contracted performance | Mostly SaaS, web and cloud traffic tolerant of internet paths |
| Trust in carrier | Carrier assurance is strong and contractually audited | You prefer to treat every transport as untrusted |
| Team capability | Small network team that relies on a managed carrier service | Team able to run and harden controllers, or a managed SD-WAN with clear shared responsibility |
| Site count and churn | Few stable sites | Many sites, frequent openings and closures |
| Regulatory or contractual demands | Requirements that specify private circuits | Requirements that specify encryption in transit regardless of carrier |
| Cloud adoption | Workloads mostly in owned data centres | Workloads mostly in cloud and SaaS |
- Small organisations: a managed or cloud-delivered SD-WAN often beats self-run controllers, provided the contract states who patches the controllers, who holds admin accounts and how logs reach you.
- Large organisations: self-managed controllers give more control and more responsibility; isolate them like identity infrastructure and keep an MPLS or private underlay where performance contracts matter.
- Selecting a product: see Security Requirements for Selecting SD-WAN.
Worked scenario
A retailer with 180 stores, two data centres and an MPLS contract up for renewal wants to move to SD-WAN over broadband with LTE backup. The security team maps traffic classes first: card payment traffic, store management, staff SaaS and guest Wi-Fi. Card payment and store management stay inside the overlay to the data centres; staff SaaS breaks out through a cloud security service; guest Wi-Fi gets its own segment with direct breakout and no route to the corporate overlay. The team chooses self-managed controllers hosted in the two data centres, reachable only from the management network and from edges, with administrator access through phishing-resistant MFA and separate roles for network and security policy. Edge certificates are issued per device and revoked when hardware is returned. Every new control-connection peer raises a ticket that must match a change record. During migration, 20 pilot stores run SD-WAN with MPLS kept as a second underlay, and the team confirms that no store can reach the internet without passing the inspection point defined for its traffic class. MPLS is retired region by region only after a store has run for a full trading month with no unexplained peers or policy drift.
Migration risk and how to manage it
- Baseline the current WAN: document routes, inspection points and which flows rely on carrier separation as their only protection.
- Harden the control plane before the first edge goes live: restrict management access, enforce MFA, set up off-box logging, and remove default or shared accounts.
- Define breakout policy per traffic class and test it in the lab, including what happens when the security service is unreachable (fail open or fail closed).
- Pilot in rings: a handful of low-risk sites, then a region, with MPLS kept as an underlay so a site can roll back by changing its preferred path.
- Watch for dual-path leaks: during coexistence, confirm that routes learned over MPLS and over the overlay do not create paths around your firewalls.
- Retire carefully: remove MPLS routes and CE configuration, revoke carrier portal accounts, and update supplier records.
- Keep a rollback plan: retain the ability to restore MPLS paths at each site until the overlay has passed the agreed run period.
Detection and measurement
- Unexplained control-plane peers: new peers without a change record (target zero).
- Management plane exposure: orchestrator and controller interfaces reachable from the internet (target zero), checked from outside.
- Administrator MFA coverage: share of fabric and carrier portal accounts using phishing-resistant MFA.
- Encrypted share of WAN traffic: proportion of inter-site traffic protected by IPsec or the overlay, by site.
- Uninspected breakout: sites or segments sending internet traffic outside the defined inspection path.
- Controller patch latency: days from supplier security advisory to patched controllers.
Send orchestrator, controller and edge logs to a central platform the fabric administrators cannot edit, and alert on software downgrades, new local accounts and SSH key changes. See the Network Device Syslog and Configuration Change Detection Guide.
Common failure patterns
- Assuming MPLS is encrypted. Sensitive flows cross the carrier core in cleartext for years because "it is a private network".
- Exposing the orchestrator for convenience. Internet-reachable management interfaces turn one supplier vulnerability into fabric-wide compromise.
- Shared admin accounts on the fabric. Without named accounts, peer and policy changes cannot be traced to a person or ticket.
- Breakout before inspection. Local breakout goes live with the first site, and the replacement for the data centre proxy arrives months later.
- Logs only on the device. Attackers in the UAT-8616 activity cleared logs on the systems they controlled; on-box logs alone cannot support an investigation.
- Forgetting the carrier after cutover. Portal accounts and residual circuits stay active long after MPLS is "retired".
Framework mapping
| Framework | Reference | Relevance to the WAN choice |
|---|---|---|
| NIST SP 800-53 Rev. 5 | SC-8, SC-8(1) | Transmission confidentiality and integrity; cryptographic protection over carrier or internet paths |
| NIST SP 800-53 Rev. 5 | SC-12 | Key establishment and management for overlay or CE-to-CE IPsec |
| NIST SP 800-53 Rev. 5 | SC-7, CM-7 | Boundary protection at breakout points; least functionality on edges and controllers |
| NIST SP 800-53 Rev. 5 | IA-3 | Device identification and authentication for edges and controller peers |
| NIST SP 800-53 Rev. 5 | SI-4, AU-6 | Monitoring and review of control-plane and management events |
| NIST SP 800-53 Rev. 5 | SR-6 | Assessment of carriers and SD-WAN suppliers |
| NIST CSF 2.0 | PR.DS-02, PR.IR-01, GV.SC-07 | Data in transit protected; networks protected from unauthorised access; supplier risk managed through the relationship |
| MITRE ATT&CK v20 | T1040, T1557, T1599, T1601 | Sniffing, adversary-in-the-middle, boundary bridging and system image modification on network devices |
Practitioner checklist
- List which WAN flows rely on carrier separation as their only protection, and encrypt the sensitive ones.
- Restrict SD-WAN orchestrator and controller management interfaces to trusted networks.
- Enforce phishing-resistant MFA and named accounts for every fabric and carrier portal administrator.
- Inventory API tokens, service accounts and edge certificates with owners and rotation dates.
- Reconcile every new control-connection peer against inventory and change records.
- Define the inspection path for each traffic class before enabling local breakout.
- Forward controller and edge logs off-box and alert on downgrades, new accounts and key changes.
- Pilot with MPLS kept as an underlay and a tested per-site rollback.
- Close carrier accounts and circuits when MPLS is retired.
Standards and references
- IETF: RFC 4364 BGP/MPLS IP Virtual Private Networks (VPNs)
- IETF: RFC 4381 Analysis of the Security of BGP/MPLS IP Virtual Private Networks
- NCSC: Using IPsec to protect data
- NIST: SP 800-77 Rev. 1 Guide to IPsec VPNs
- NIST: SP 800-53 Rev. 5 Security and Privacy Controls
- CISA: Known Exploited Vulnerabilities Catalog
- Cisco Talos: UAT-8616 exploitation of Cisco Catalyst SD-WAN
Related NHI Mgmt Group resources: SD-WAN Security Guide · MPLS and Carrier WAN Security Guide · Security Requirements for Selecting SD-WAN · IPsec Security Guide · Edge Device Security Guide