FreeRDP CVE-2026-121394

Exploitation requires the victim to initiate an RDP connection to a server under the attacker's control (or sit behind a relay/MITM) AND to have RemoteFX surface decoding enabled in the FreeRDP client (the PoC uses +rfx). It is further gated by build configuration: the process-terminating abort only occurs when the client is compiled with WITH_VERBOSE_WINPR_ASSERT enabled - the default in the inspected FreeRDP 3.32.1 CMake config, but some project presets and downstream distribution builds disable it, in which case the narrowing conversion does not assert and the client remains alive. The advisory's /sec:rdp usage is a test convenience and is NOT a prerequisite. These limits (required outbound connection to a hostile server, RemoteFX enabled, assert-enabled build) narrow realistic exposure considerably below the raw network/unauthenticated metrics.

This is a client-side denial-of-service with no confidentiality or integrity impact, which caps real-world risk well below what the 7.1 headline might suggest. The input CVSS 4.0 vector (AV:N/AC:L/AT:N/PR:N/UI:P/VC:N/VI:N/VA:H) tells a consistent story: network-reachable and unauthenticated against the vulnerable code path, but gated by passive user interaction (UI:P) - the victim must actively connect to a server the attacker controls, so there is no server-side or wormable exposure and no path to code execution. The vuln.today decision label is 'this_week' (not act_now/emergency). There is no CISA KEV listing and no EPSS score in the provided data, so there is no signal of active or widespread exploitation - appropriate for a crash-only client bug. A key mitigating nuance the signals do not capture: the GHSA advisory states the abort is build-dependent, triggering only when WITH_VERBOSE_WINPR_ASSERT is compiled in (default ON in the inspected 3.32.1 config, but disabled in some presets and downstream builds, where the conversion does not assert and the client stays alive). The CVSS vector does not encode this configuration dependency. Net: a genuine but limited priority - worth patching on a normal cadence for anyone running FreeRDP-based clients against untrusted servers, not an emergency.

An attacker stands up or compromises an RDP server (for example, luring a user via a crafted .rdp file, a malicious connection broker, or a man-in-the-middle relay as demonstrated in the advisory's PoC) and waits for a FreeRDP-based client with RemoteFX surface decoding enabled to connect. Once connected, the server sends a RemoteFX tileset whose RLGR bitstream emits code 32767, triggering a signed narrowing assertion in rfx_rlgr_decode and aborting the client process. A public proof-of-concept exists in the GHSA advisory (a loopback relay script with recorded crash runs showing exit 134); the input does not mark it as confirmed weaponized exploit code, and the impact is limited to crashing that one client process.