Quiet Windows password attack gets a cracker and detections

Published

A Windows domain controller will hand anyone on the network a password-derived fingerprint for almost any computer account, without asking who they are. That is timeroasting, and new research from Andrew Schwartz at Huntress shows that the version of it most defenders and tools ignore is live on a fully patched Windows Server 2025 domain controller, crackable, and poorly covered by today's detections.

The attack itself is not new. Tom Tervoort published it in 2023: Microsoft's time-sync extension, MS-SNTP, lets a client name an account by its Relative Identifier (RID, the account's index number in the domain) and the domain controller answers with a cryptographic checksum keyed on that account's password hash. An attacker sweeps RIDs, collects the checksums and cracks them offline. Because the targeted accounts never log on, there is no 4768, 4769, 4624 or 4776 event for logon-based analytics to catch.

The 120-byte form nobody was cracking

MS-SNTP answers in two formats. The familiar one is a 68-byte packet whose checksum is built with MD5, which hashcat mode 31300, timeroast.py and NetExec all handle. The other is a 120-byte extended form that Microsoft added to the specification as the preferred option because of MD5's weaknesses, built from an SP800-108 key derivation step followed by HMAC-SHA512. Modern Windows clients send it by default, but its exact byte encoding was not fully documented and no released hashcat build can crack it.

Schwartz reverse-engineered the time service and Netlogon binaries from a patched Windows Server 2025 domain controller. The time service turned out to hand the signing work to Netlogon, where a function named NlGetKDFKey pins down the details the specification leaves vague: an 8-byte label that includes its trailing null, a little-endian RID as context, and mixed endianness inside a single buffer. Probing a live domain controller, he found the extended form answered for the account's previous key, and a reference implementation reproduced the controller's checksum exactly from the account's hash.

He has also written a hashcat module under the placeholder mode 31350, which passes hashcat's own self-test and benchmarks at 17.6 MH/s on an Apple M3 Pro. It is not yet upstream. The extended form is much slower to crack per guess, but Schwartz is clear that this barely matters: both forms key off the same password hash, so a weak password falls to either and a random machine secret falls to neither.

Why packet size is the wrong signal

An attacker has a reason to prefer the extended form even though it is harder to crack: on networks where legitimate clients already send 120-byte requests, it blends in. Schwartz warns against alerting on request size, and against alerting on an all-zero checksum, because a legitimate domain member sends exactly the same thing. In his 11 test runs over five hours, Microsoft Defender for Identity raised no timeroasting alert.

The detection he calls durable is an own-RID check, run in Zeek or a SIEM rather than as a Suricata signature: for each authenticated time request to a domain controller, read the requested RID, resolve the source address to its computer account, and alert when a machine asks for a RID that is not its own. The post also ships a simpler Suricata rule that flags 25 authenticated requests from one source within 60 seconds, which it describes as weaker because a paced attacker can stay under it, and a rule for the old-key flag that Schwartz labels a hypothesis to baseline first. EDR telemetry showing a non-time-service process sending UDP/123 to a domain controller is another strong signal.

Lock down UDP/123 and rotate machine passwords

Huntress recommends limiting UDP/123 access to domain controllers to the networks that need it, keeping automatic machine-account password rotation on, fixing predictable pre-Windows 2000 computer-account passwords, and monitoring authenticated time requests for unexpected sources and RIDs. For organizations that already hardened their directory after this year's domain controller takeover bugs, the lesson is the same one in a quieter place: timeroasting never touches a logon event, so the only place to see it is the network traffic most teams are not reading.

This briefing is provided by IntelFusions for informational and defensive purposes only. It is based on sources assessed to be reliable at the time of writing, and analytic judgments carry the confidence levels indicated. Indicators of compromise are defanged; re-arm them only in controlled environments. IntelFusions is not affiliated with the organizations named and makes no warranty as to completeness or accuracy.

Earlier in Vulnerabilities

All Vulnerabilities briefings

Latest briefings

Read the full analysis on IntelFusions