BGP hijacking: how it works, real cases and RPKI protection

What kind of hijacking are we talking about?
BGP hijacking happens when a network operator announces IP address ranges that don't belong to it. Part of the world's traffic then flows to that operator instead of the legitimate owner.
The word "hijacking" covers three different things in IT:
- session hijacking: stealing a cookie or token to impersonate a logged-in user;
- domain hijacking: taking control of a domain name, often by compromising the registrar account;
- BGP hijacking: diverting routes, which attacks the infrastructure of the internet itself.
This article covers only the last of these: how it works, dated incidents, protection with RPKI, and what it means for a VPS or dedicated server.
In short
- BGP is built on trust: nothing natively checks that an operator is entitled to announce a prefix.
- A more specific prefix (a /24 against a /22) wins everywhere it propagates.
- RPKI and route origin validation (ROV) block announcements with a spoofed origin, but not those with a forged AS path.
- Multi-homing and local peering reduce the reach of a hijack without preventing it.
- Encryption (TLS, with HSTS) and domain-level controls (DNSSEC, CAA) remain your protection on the client side.
What is BGP hijacking?
BGP (Border Gateway Protocol, version 4 is described in RFC 4271) is the protocol networks use to exchange routes with each other. Each network is an autonomous system (AS) identified by a number, and it announces prefixes: blocks of addresses in CIDR notation, such as 203.0.113.0/24. The global routing table contains tens of thousands of ASes.
The weakness is historical. An announcement is accepted on the word of whoever sends it, and no built-in check confirms that the AS has the right to announce that prefix.
Three situations can occur:
- a malicious, deliberate hijack;
- a configuration error (the "fat finger"), which has the same effect without any intent;
- a route leak, defined in RFC 7908: legitimate routes are propagated beyond their intended scope, for example when a customer re-announces them to another provider.
How a BGP prefix hijack works
Take the documentation range from RFC 5737. A hosting provider (AS64496) legitimately announces 203.0.112.0/22, and an attacker (AS64511) wants to divert the traffic of a server at 203.0.113.10.
Announcing a more specific prefix
Routers apply longest prefix match: when two routes cover the same address, they pick the more precise one. The attacker therefore announces 203.0.113.0/24. Every router that receives it prefers that /24 over the legitimate /22, whatever the path length. It is the most effective scenario.
Origin and path spoofing
The simplest variant announces the same prefix with a shorter AS path, to attract routers that compare paths.
The subtler variant is called a forged-origin hijack. The attacker builds a path "64511 64496" in which the legitimate AS appears last. The origin looks right, so it passes an origin check. The path, however, is false.
The consequences depend on intent:
- blackholing: traffic is dropped and the service becomes unreachable;
- interception: the attacker reads or alters the traffic, then forwards it to its destination (a man-in-the-middle attack);
- cryptocurrency theft through a fake website;
- fraudulent TLS certificates obtained through domain validation, since the attacker receives the verification traffic;
- spam sent from stolen IP ranges.
The best-known BGP hijacking incidents
| Date | Actor | Target | Duration | Impact |
|---|---|---|---|---|
| 24 February 2008 | Pakistan Telecom (AS17557) | YouTube, 208.65.153.0/24 | about 2 hours | Part of YouTube's traffic diverted |
| 8 April 2010 | Chinese operators (including China Telecom) | About 15% of global traffic | about 18 minutes | US government sites affected |
| 24 April 2018 | eNet (AS10297) | DNS service of a major cloud provider | about 2 hours | MyEtherWallet users redirected, ETH stolen |
| 1 April 2020 | Rostelecom | Many prefixes | not documented here | Several major online services affected |
Here is each case in turn, with its source.
- YouTube (2008). According to the RIPE NCC case study, the /24 announcement began at 18:47 UTC. An upstream provider removed the operator's routes at 21:01. YouTube was already announcing a /22, but the more precise /24 won.
- 2010. ABC News, relaying a US report, mentions around 15% of global traffic being diverted for about 18 minutes.
- Cloud DNS service (2018). The more specific routes 205.251.192.0/24, .193.0/24 and .195.0/24 were announced from 11:05 to 13:03 UTC. The rogue DNS pointed to a copy of the crypto wallet site, with a valid TLS certificate issued by Let's Encrypt, so browsers showed no warning. The exact amount stolen is not given here.
- Rostelecom. ThousandEyes describes more specific routes announced across a large number of prefixes. We give no timings or counts here.
One confusion comes up often: Facebook's outage of 4 October 2021 was not a hijack. Facebook withdrew the routes to its own DNS servers, and its services vanished. That is the opposite of a hijack: nobody stole those prefixes, they simply stopped being announced.
RPKI and ROAs: the reference protection
RPKI (Resource Public Key Infrastructure, RFC 6480) relies on certificates issued by the regional internet registries: the RIPE NCC for Europe, the UK included. The holder of an IP block publishes a ROA (Route Origin Authorisation) there, a signed object that binds three things: a prefix, the AS allowed to announce it and a maximum length (maxLength).
Operators then apply route origin validation (RFC 6811). Each announcement gets a state:
- valid: a matching ROA exists;
- invalid: a ROA exists, but the origin AS or the length doesn't match;
- not found: no ROA covers the prefix.
A "drop invalid" policy rejects invalid routes. In our example, with a ROA limited to the /22, the attacker's /24 is dropped.
Coverage grows every year, but figures vary by source and by protocol (IPv4 or IPv6). Check a current RPKI measurement dashboard rather than relying on a number that will date quickly. The MANRS initiative, run by the Internet Society, brings together operators committed to good routing-filter practice.
What RPKI doesn't cover yet
RPKI validates the origin, not the path. Forged-origin hijacks and route leaks slip past it. BGPsec (RFC 8205) signs the path, but it is barely deployed. ASPA, meant to detect abnormal paths, is still being standardised at the IETF, so check its status before counting on it.
Two habits limit the damage. First, keep a strict maxLength, because an overly broad ROA lets a spoofed sub-prefix through. Second, encrypt end to end with TLS and HSTS, so that interception yields nothing. DNSSEC, for its part, lets validating resolvers spot forged DNS responses, as in the 2018 case.
Multi-homing and peering: reducing the impact on your servers
A multi-homed host relies on several upstream providers. If it is also present at an exchange point such as France-IX, it has direct, short routes to French ISPs. A rogue announcement coming from far away is less likely to be preferred, and it spreads less well along those paths.
To be clear, multi-homing does not block a hijack. It reduces its reach and speeds up remediation. That means monitoring prefix origins, announcing more specific prefixes (YouTube did so in 2008) and contacting upstreams quickly. To judge how seriously a host takes this, ask about its routing setup, its upstreams and its monitoring.
Don't confuse this with a DDoS attack either: DDoS overwhelms a target, whereas a hijack redirects its traffic. For that other risk, the same host should be able to explain its filtering capacity and how it handles attacks. Our piece on intrusion prevention systems versus IDS and firewalls covers the host-level defences that complement network filtering.
On the client side, the checklist comes down to five points:
- HTTPS everywhere, with HSTS;
- DNSSEC on your domains;
- CAA records to limit which certificate authorities may issue for you;
- availability monitoring from several regions;
- a host that documents its network, such as a Linux VPS hosted in Paris.
Frequently asked questions
What is the difference between BGP hijacking and session hijacking?
Session hijacking steals an application session, a cookie for example, to impersonate a user. BGP hijacking works lower down: it diverts routes between operators, and therefore the traffic of every user of a prefix. The first targets an account, the second a range of IP addresses.
Can BGP hijacking affect a simple VPS?
Yes, indirectly. The attacker diverts a whole prefix, not a single machine, so every server in the range is affected. Targeted attacks mostly go after high-value services such as crypto, but a broad hijack hits every service in the range, VPSs included.
Is RPKI enough to prevent a BGP hijack?
No. It blocks announcements with an illegitimate origin, provided operators apply validation. It doesn't check the AS path, so a forged-origin hijack or a route leak can get through. Combine it with a strict maxLength, monitoring and end-to-end encryption.
How do I know whether my IP addresses have been hijacked?
Monitor the origin of your prefixes with BGP monitoring tools, which alert you to any change of origin AS or any more specific announcement. On the client side, availability probes from several regions and monitoring of certificates issued for your domains reveal symptoms. Ask your host as well.
Does HTTPS protect against BGP hijacking?
Largely, as long as the certificate is legitimate: an attacker who intercepts the traffic can't read it. They may, however, try to obtain a fraudulent certificate through domain validation. HSTS, CAA and DNSSEC reduce that risk.
What is the difference between a route leak and a BGP hijack?
A hijack announces prefixes the sender doesn't own. A route leak propagates legitimate routes beyond their intended scope, often through a configuration error, as RFC 7908 describes. Both divert traffic, but the origin of the announcement differs.
Where should you start with your server?
Start with what you control: enable HSTS, switch on DNSSEC and add a CAA record to your domains. Then ask your host how it manages its routes, its upstream providers and its monitoring.
At OmegaCloud, VPSs and dedicated servers are hosted at Telehouse Paris, on a 10 Gbit/s network that is multi-homed over BGP, with France-IX peering. 1 Tbit/s Anti-DDoS protection is included and human support answers 24/7. The offers are currently available for pre-order, with availability announced for 1 November 2026.