Read the docs

Remote Credential Guard leaks passwords to RDP server in FreeRDP

Priority MEDIUM 44CWE-522Published · Updated

Key facts

Attack vector
Network
Privileges required
None
User interaction
Required
Attack complexity
Low
Impact type
Information disclosure
Confidentiality · integrity · availability
High · None · None
Fix status
Fixed

Attack profile from CVSS 3.1 · GHSA-hhfh-fvm7-7w65. Fix status from GitHub. Impact type from CWE-522.

Summary

FreeRDP libfreerdp 3.32.2 sends a configured password to the RDP server in the Client Info packet when Remote Credential Guard is used. The Client Info code suppresses credentials for Restricted Admin and disabled delegation but not for Remote Credential Guard, which is a missing check (CWE-522). A malicious or impersonated RDP server that supports Remote Credential Guard can read the password from a client that connects with a configured password.

Written with Claude Haiku 5.5 from the sources below · 2026-11-12

Description from the GitHub advisory

### Summary FreeRDP 3.32.2 suppresses Client Info credentials for Restricted Admin and disabled delegation, but omits Remote Credential Guard (RCG) from that guard. A client connecting with /sec:nla +remoteGuard and an already configured /p password completes Kerberos/CredSSP with TSRemoteGuardCreds, then sends that password in the subsequent Client Info packet. In a bounded 3.32.2 loopback test, a controlled RCG-capable server parsed the fixed synthetic password lab from the unchanged client and reached ACTIVE; an otherwise matched no-/p control parsed an empty password field. This is an RCG variant of the Client Info sink already disclosed for Restricted Admin in GHSA-jhxw-3hj9-9hqh. ### Details `libfreerdp/core/info.c::rdp_write_info_packet()` serializes domain, username and password unless `RestrictedAdminModeRequired` or `DisableCredentialsDelegation` is set. It does not check `RemoteCredentialGuard`. The `/remoteGuard` option and `/p` are independently accepted by the 3.32.2 client. The normal connection path calls the writer after Kerberos/CredSSP/RCG, so `TSCREDS_REMOTEGUARD` alone does not prevent the later Client Info disclosure. Exact 3.32.2 source commit: `97eff6ddfbbfaa8e42a126d362686bd001ff8bdc`. The client binary and library were hash-gated and unchanged in the paired run. Official master `dae0390a42ab271fa07a35979dfcd5671f83a949` retained the same omitted RCG guard on 31 October 2026 by source review. The complete affected-release range is not yet determined. The loopback peer was FreeRDP Shadow built separately with a **test-only** guard for an omitted optional CredSSP `supplementalCreds` field; stock Shadow otherwise aborts this valid RCG handshake. An observer after production Client Info parsing logged only `cbPassword` and a boolean equality test against fixed `lab`. This proves the unchanged client's endpoint disclosure to an RCG-capable peer, not a stock Windows or unmodified FreeRDP-server run. The full report, source patch, bounded runner and minimized paired receipts are in the verified D-512 attachment. No real credential was used. Please assess this as an incomplete fix or related variant of GHSA-jhxw-3hj9-9hqh, which publicly names the same Client Info sink for Restricted Admin and lists 3.32.0 as patched. Its published text does not discuss RCG. #### Related secondary mechanism: D-512-B (separate negotiation root) The 3.32.2 `nego_process_negotiation_response()` path accepts a peer's RCG-support assertion without requiring CredSSP as the selected protocol, then checks a selected protocol against globally enabled modes rather than the current request mask. [MS-RDPBCGR's RCG-required flag](https://learn.microsoft.com/en-us/openspecs/windows_protocols/ms-rdpbcgr/902b090b-9cb3-4efc-92bf-ee13373371e3) requires credential-less logon over CredSSP; its [negotiation sequence](https://learn.microsoft.com/en-us/openspecs/windows_protocols/ms-rdpbcgr/db98be23-733a-4fd2-b086-002cd2ba02e5) selects a requested protocol or returns failure. In the separate bounded full-client observation, an RCG-required request offered `0x0b` (SSL/HYBRID/HYBRID_EX), but the controlled peer selected classic RDP (`0`) while asserting RCG support. The client accepted this unoffered selection, skipped CredSSP/`TSRemoteGuardCreds`, and sent a 331-byte Client Info follow-up whose declared password field contained the fixed synthetic `lab` value (six UTF-16LE bytes). A matched `/remoteGuard /sec:nla` negative control offered `0x03` (SSL/HYBRID), rejected the same classic selection and sent zero follow-up bytes. The attachment includes all six minimized packets and an independent semantic verifier. This is an RCG-*requested classic-RDP downgrade*, not the completed normal Kerberos/CredSSP/RCG session that supports the primary claim above. The negotiation root is distinct from the Client Info writer defect, while the demonstrated password sink overlaps the same public advisory family. Please assess its scope alongside the primary RCG variant. An independently reasoned **provisional** CVSS v3.1 6.5 (`CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:N/A:N`) applies only to this observed downgrade with an already configured reusable password: the destination controls the deterministic selection, a user must connect, and the complete password reaches the peer. The same number as the main claim is coincidental, not additive. ### Proof of concept Use only the attachment's `tools/d512-rcg-full-session/README.md` inside an isolated lab. The Dockerfile derives from the pinned 3.32.2 Kerberos/RDPEAR client image, checks source/client hashes, preserves the client binary, and builds the test-instrumented Shadow server separately. Run one no-password control and one `/p:lab` observation with `run_container.sh`, each under its own tracked 60-second, no-network loopback container. The runner requires HYBRID negotiation, `TSRemoteGuardPackageCred(Kerberos)`, server ACTIVE and the post-parse Client Info marker. Control: `cbPassword=0 sentinel_match=0`. Observation: `cbPassword=6 sentinel_match=1`. Run `verify_receipts.py` over the packaged positive and negative synthetic consoles. It checks the recorded marker order, case-specific values, exits and log hashes; it does not rerun the target. The attachment also preserves the earlier exact writer pair and a separate classic-downgrade packet pair. Raw logs, Kerberos tickets, keys and real credentials are excluded. For the separate D-512-B mechanism, run `verify_receipt.py` from `package/lab/tools/d512-rcg-session/` on `package/lab/receipts/observation` with `--expect-info`, then on `package/lab/receipts/nla_only_control` without that flag. The first checks the `0x0b` request, classic selection and 331-byte Client Info containing synthetic `lab`; the second checks the `0x03` request and zero follow-up. These validate recorded packet evidence, not a new target run or a completed RCG authentication. ### Impact A malicious or compromised RCG-capable destination receives an explicitly configured reusable client password that RCG is intended to withhold. The tested positive required the victim to connect using RCG and an already populated /p password. Ticket-only RCG without a password did not expose one. D-512-B demonstrates a separate RCG-requested classic-RDP downgrade to the same Client Info password sink; its conditional score is assessed separately and adds no impact to this primary claim. No password reuse, post-theft action, stock-Windows-server behavior, authentication bypass, integrity effect or code execution was tested.

Why it matters now

Fix Fixed libfreerdp: 3.33.0

SSVC decision points

Not available: CISA has not published SSVC decision points for this record.

Affected products

VendorProductAffected versionsFixed inStatusSource
—libfreerdp3.32.23.33.0Patch availableGHSA-hhfh-fvm7-7w65
  • libfreerdp Vendor— Affected 3.32.2 Fixed in 3.33.0 Patch available GHSA-hhfh-fvm7-7w65

Remediation

Upgrade libfreerdp to 3.33.0, the fixed version listed for the product.

Written with Claude Haiku 5.5 from the sources below · 2026-11-12

  • Fix · GHSA-hhfh-fvm7-7w65 libfreerdp Fixed in 3.33.0 Advisory →

Timeline

  1. 2026-11-12 Threatist record created from GHSA-hhfh-fvm7-7w65

References

Sources

This product uses the NVD API but is not endorsed or certified by the NVD. Sources and licences