What Is Secure Remote Password (SRP)?
Secure Remote Password (SRP) lets a client prove a password without sending it. Learn how SRP works and why a key leak broke vCenter authentication.
Secure Remote Password (SRP) is a password-based authentication and key-exchange protocol that lets a client prove it knows a password without ever sending the password, or anything an eavesdropper could replay. Both sides finish the exchange holding the same fresh session key. The design assumes an attacker can read and modify every message, so the only secret that matters is the one the user memorized. When an implementation leaks that session key, the protocol's guarantees fall apart, which is the root of vCenter CVE-2026-123890.
What is the Secure Remote Password protocol?
Secure Remote Password is an authentication protocol in which the server stores a verifier derived from the password instead of the password itself, and the two parties compute a shared session key through a mathematical exchange. Stealing the verifier does not let an attacker impersonate the user, and nothing sent on the wire can be replayed. It was first proposed in 1996 and published as SRP-3 in 1998. SRP-6 and SRP-6a are the variants in common use today.
The Stanford SRP project, which maintains the protocol, lists the conditions it is built to withstand: attackers know the protocol in full, hold a large dictionary of common passwords, can eavesdrop on every message, can intercept and forge traffic, and have no trusted third party to lean on. Most password schemes fail at least one of those tests. SRP is designed to pass all of them, which is why it appears inside enterprise products that need machine-to-machine authentication over internal channels.
How does SRP authenticate without sending a password?
SRP never transmits the password. The client and server each combine the password material they hold with random values generated for that one session, and both arrive at the same session key only if the client's password matches the server's stored verifier. Each side then proves it holds the key with a short message, and the key protects the rest of the conversation.
Three properties fall out of that design:
- No replay. The random values change every session, so a captured exchange is useless the next day.
- No offline guessing from traffic. An eavesdropper cannot test password guesses against what they recorded.
- Verifier theft is not password theft. The server-side verifier lets the server check the client, but it does not let a thief log in as the client.
That third property is why a protocol like SRP is preferred over storing hashed passwords that get compared in the clear. The server holds something strictly weaker than the password, which is why SRP is called an augmented password-authenticated key exchange.
Where does vCenter use SRP?
vCenter Server runs an internal directory, the VMware Directory Service (vmdir), that stores Single Sign-On accounts, the SAML signing material for the Security Token Service, and machine credentials. Other vCenter components authenticate to vmdir, and that internal channel uses SRP-6a. Mandiant's write-up on the ARMORY CASTLE campaign identifies the affected service as the SRP verifier endpoint reachable on TCP port 2012, which it describes as the control interface for vCenter Single Sign-On.
That service is not meant for people. Administrators log in through the web client over HTTPS on port 443. The SRP channel exists so vCenter's own services can reach the directory, which is exactly why it carries high trust and why it is easy to overlook.
Why does a session key leak defeat the protocol?
SRP's security rests on one assumption: the session key is known only to the two legitimate parties, because computing it normally requires the password or the verifier. A disclosure flaw breaks that assumption. If an unauthenticated caller can obtain the shared session key for a privileged account, the server treats the rest of the exchange as genuine, and the caller holds an authenticated session it never earned. In CVE-2026-123890 the recovered session belonged to administrator@vsphere.local, the built-in vCenter super-administrator. No password was cracked and no verifier was stolen. The key that was supposed to stay secret did not.
The lesson for defenders is that a strong protocol is only as good as the implementation underneath it. SRP is sound. The flaw was in how one product's service handled the key material around it. Two takeaways follow. First, internal authentication channels deserve the same patch urgency as the login page, because they often carry more privilege. Second, network reachability is the real control: a service that implements SRP perfectly still should not be reachable from a general-purpose server, because an implementation bug can turn that reachability into full access. That is why the response to this vulnerability pairs patching with auditing vCenter SSO accounts after a compromise and with limiting what can reach the vCenter appliance in the first place.
For the full timeline, affected versions and detection steps, see our news report on the exploited vCenter authentication bypass CVE-2026-123890.