Reference · Mechanism
How a fake base station works, and why it still works
Nothing about the technique is subtle. It works because a handset must choose a cell before it can check anything, and for three generations there was nothing for it to check.
Published
1Asymmetry
Authentication that ran in only one direction
The behaviour at the centre of this is a design decision from the era from the era when mobile networks were being built to work. A handset arriving in a new area surveys what it can hear, picks the cell that offers the best service, and asks to attach. The network then verifies the subscription, using the key held on the subscriber module, and admits it.
What that sequence lacks is the mirror image. The network establishes that the subscription is genuine; the handset establishes nothing whatever about the network. In the second generation this was explicit and unremedied, and a device that broadcast the right identifiers and a strong enough signal would be selected on its merits, because signal strength was the merit being assessed. Later generations added mutual authentication, which closes the impersonation of an established session, but the disclosure being discussed here happens earlier than that: before there is a session to protect.
This is why the technique is not really an exploit in the usual sense. It does not depend on a flaw in an implementation, a weak key or a protocol error. It depends on the handset doing what it is supposed to do, at the one moment when it has no basis for doing anything else. Whether a device using that opening is operating under an authorisation is a separate question, and one the mechanism cannot answer: see the comparison with interception inside the network.
2Disclosure
Why the identity is available at all
Networks were aware that transmitting a permanent identity in the open was undesirable, and they addressed it with temporary identifiers: once a subscription is known, it is addressed by a short-lived value instead, and the permanent identity stays off the air. In steady state this works.
The weakness is in the fallback. There has to be a way to establish the permanent identity in the first place, and there has to be a way to recover when the temporary value is unknown to the network, has expired, or belongs to a different network. So the protocol retains a message that asks the handset for its permanent identity, and the handset answers it, because a handset that refused would be unable to attach anywhere new.
An interception device therefore does not need to defeat anything. It needs to be selected, and then to be in a position where asking the ordinary question is reasonable. Both halves are available to it: the first because cell selection favours the strongest candidate, the second because a network that has never seen a given subscription before is exactly the situation the fallback exists for.
A related move makes the position easier still. The older the generation, the weaker the protections that follow the initial exchange, so a device that can persuade a handset to prefer an older technology inherits a simpler problem. That preference is influenceable, and the resulting drop is the signal detection leans on most heavily, precisely because an attacker has a use for it.
3Closure
What the fifth generation changed, and what it left alone
The countermeasure does not try to make cell selection safe, which would mean solving the problem of trusting a stranger before any relationship exists. It removes the prize instead. The permanent identifier is encrypted on the device, using a key belonging to the home operator, before it is ever transmitted, so the ordinary question can still be asked and the ordinary answer is no longer worth collecting: it cannot be resolved to a subscription by anyone but the home operator, and it does not repeat, so two sightings cannot be matched to each other. The concealment page sets out the mechanism and its limits.
Three things are deliberately left alone. Cell selection still prefers the best signal, so a handset can still be induced to attach to a device that is not what it claims. The routing prefix that says which operator to ask remains readable, because the network that receives the request has to know where to send it. And the concealment applies to the subscriber identity and not to the equipment identifier, which is a distinction with consequences for anyone relying on a change of subscription as a precaution.
DRecords
Specifications in this archive that bear on it
The archive’s contribution to this question was on the disclosure: protecting the identity at the moment it would otherwise be handed over, and separating a device from the identity it presents.
| Cluster | Specification | Concern | Pages |
|---|---|---|---|
| T3.2 | Privacy-enhanced identity protection | Concealment of subscriber identifiers | 13 |
| T3.2 | Device-based anonymization | Anonymization performed at the device | 10 |
| T3.1 | IoT group AKA | Group authentication for constrained devices | 13 |
QQuestions
Questions and answers
Why does a handset trust a cell it has never seen?
Because on the generations where this problem originates it has no way not to. Authentication in those designs runs in one direction: the network satisfies itself that the subscription is genuine, and the handset has no equivalent means of satisfying itself that the network is. Selecting the strongest suitable cell is therefore the only available policy, and it is the correct one for the purpose it was designed for, which was coverage rather than resistance to impersonation.
Does it have to break encryption?
No, and that is what makes the attack durable. The identity is disclosed during the exchange that precedes protection being established, so nothing has to be decrypted. An interception device that only ever collects identities never encounters encryption at all.
What stops it on a 5G network?
The permanent identifier is encrypted to the home operator before it leaves the device, so the request that used to yield a usable identity now yields a value that cannot be tied to a subscription or matched against a previous sighting. The behaviour being abused, a handset attaching to whichever cell presents itself best, is unchanged; what changed is that the attachment no longer discloses anything durable.
Is it still a live problem in 2026?
On older generations, yes, and those generations are still carrying traffic. A handset that can be induced to use second-generation service is exposed in the way it always was, and coverage gaps make that inducement plausible in ordinary places. The protection is real but it applies where the newer generation applies.