Many organizations publish ROAs, see their prefixes marked "valid" in monitoring, and consider BGP hijacking handled. Route Origin Validation (ROV) is worth deploying, but it answers one question: is this ASN authorized to originate this prefix? It says nothing about the path the route took to reach you. Below are the attack classes ROV cannot see and the controls that cover them.
What ROV actually checks
A ROA binds three things: a prefix, a maximum length, and an origin ASN.
ROA: 203.0.113.0/24 maxLength 24 origin AS64500
A validating router (RFC 6811) compares the rightmost ASN in the AS_PATH with the ROAs covering the prefix. The result is one of three states:
- Valid: a ROA matches prefix, length, and origin.
- Invalid: a ROA covers the prefix, but the origin or length does not match.
- NotFound: no ROA covers it.
Nothing else in the AS_PATH is examined, and that gap is what the rest of this article is about.
Scenario 1: forged-origin path
AS64500 owns 203.0.113.0/24 and has a correct ROA. An attacker at AS64666 announces the same prefix with a fabricated path that ends in the legitimate origin:
Legitimate: 203.0.113.0/24 path: 64510 64500
Forged: 203.0.113.0/24 path: 64666 64500
The rightmost ASN is 64500, so ROV marks the route valid. The attacker does not need to be adjacent to AS64500 or hold any credential for it. The route is one hop longer than a direct announcement, but in many topologies the attacker's neighbors still prefer it, because of local-preference policy or a shorter path from their vantage point.
A sloppy ROA makes this worse. A ROA with maxLength 24 on a /22 authorizes every more-specific down to /24. A forged-origin announcement of a /24 inside that range validates, and longest-prefix-match then pulls traffic to the attacker regardless of path length. Loose maxLength turns a path-forgery weakness into a reliable hijack.
Scenario 2: the route leak
A leak needs no forgery at all. In the classic case, a customer AS receives routes from one provider and re-announces them to another, which violates the valley-free model (RFC 7908). Every leaked route keeps its genuine origin, so every one validates under ROV.
The 2019 incident in which a small network leaked tens of thousands of prefixes through a large transit provider followed this pattern. Major content and CDN prefixes were among those leaked, and traffic was drawn through an under-provisioned path. Every origin was correct, so ROV passed them all.
What it costs
- Traffic blackholing or degradation. Leaked traffic can be dropped or squeezed through a path that cannot carry it. Outages in these incidents ran from tens of minutes to hours.
- Interception. A forged path can place an attacker on-path for inspection. TLS limits the damage, but not for unencrypted traffic, DNS, or certificate issuance flows.
- Compliance exposure. An auditor who sees "RPKI deployed" and treats routing integrity as closed has an incomplete picture. Reviews against frameworks such as MANRS or NIS2-style risk obligations will ask what covers path integrity.
- Misleading assurance. A dashboard showing 100% valid routes during an active leak is worse than no dashboard, because it suppresses escalation.
Where ASPA and BGPsec fit
ASPA (Autonomous System Provider Authorization) targets leaks and many forged paths. An AS publishes a signed object listing the providers it buys transit from:
ASPA: customer AS64500 providers { AS64510, AS64511 }
A verifier walks the AS_PATH and checks each customer-to-provider hop against these objects. If a customer's route goes "up" to an AS that is not its authorized provider after having gone "down", the path is flagged invalid. This catches classic leaks and many forged-path announcements without per-hop signing. It has limits. It covers only ASes that publish ASPA objects, and it infers direction from published relationships. The verification procedure is also still being finalized in the IETF SIDROPS working group, so check current router support before relying on it.
BGPsec (RFC 8205) cryptographically signs each hop, so the path itself is attested. In theory it prevents path forgery outright. In practice it needs every AS on the path to participate, the signatures add significant processing overhead, and deployment is negligible. Treat it as a long-term option, not a mitigation you can adopt now.
BGP Roles and the OTC attribute (RFC 9234) are a lighter control available today. Neighbors declare their relationships, and a route learned from a provider or peer is marked with an Only-To-Customer attribute so it cannot be re-announced upward.
How monitoring catches it
ROV state is blind to these attacks, so detection has to look at paths, not just origins.
- Path-aware alerting. Compare observed AS_PATHs for your prefixes against an expected set of upstream neighbors. An unexpected AS adjacent to your origin is the signal for forged-origin attacks.
- Public route collectors. RIPE RIS and RouteViews data show how the global table sees your prefixes. BGPalerter can alert on new upstreams:
# bgpalerter prefixes.yml
203.0.113.0/24:
description: primary service
asn: 64500
ignoreMorespecifics: false
group: core
# config.yml (monitor)
- file: monitorNewPrefix
- file: monitorPath # alerts on paths not matching expected patterns
path:
- match: ".*64500$"
notMatch: ".* (64510|64511) 64500$"
matchDescription: unexpected upstream before our origin
- Leak-specific signals. Watch for your routes appearing through unexpected transit ASes, and for sudden latency or traffic-volume shifts on ingress.
- maxLength audit. Compare announced prefix lengths with ROA maxLength values. Any gap is a forged-origin more-specific waiting to happen.
Practical mitigation steps
- Tighten ROAs. Set maxLength equal to the announced length, and publish separate ROAs for each more-specific you actually announce. Remove stale ROAs.
- Keep ROV and add filtering. Reject invalids, and also filter customer sessions by prefix lists built from IRR/RPKI data:
bgpq4 -4 -b -l CUST-ACME AS-ACME # generate customer prefix-list
- Limit the blast radius. Apply per-session
max-prefixlimits so a leaking customer session shuts down rather than flooding your table. - Enable RFC 9234 roles where your router software supports them (BIRD, FRR, and OpenBGPD do), so leak prevention is enforced by default.
- Publish ASPA objects for your ASN through your RIR's RPKI portal once available, and enable ASPA verification as router implementations mature.
- Agree on a response path. Keep NOC contacts for your upstreams and peers, and a runbook for withdrawing, prepending, or announcing more-specifics during an event.
- Re-check regularly. SetEnforce's RPKI checks can flag missing ROAs and loose maxLength values, but path-level protection needs the monitoring and filtering controls above.
Key takeaways
- ROV validates only the origin ASN. A forged path ending in your ASN and a leak of your genuine routes both report "valid."
- Loose maxLength lets forged-origin more-specifics succeed. Keep it exact.
- ASPA is the practical next layer for leaks and upstream path anomalies. BGPsec is theoretically complete but unrealistic today.
- RFC 9234 roles and OTC, prefix filters, and max-prefix limits give you leak protection now.
- Monitor AS paths and upstream adjacency, not just validation state, and do not cite "all routes valid" as evidence of routing integrity.