Tencent HunyuanDiT CVE-2025-13712

Exploitation requires a user running Tencent HunyuanDiT to load or process an attacker-supplied artifact through the merge endpoint - the condition is that untrusted, attacker-controlled serialized data reaches that endpoint, where it is deserialized without proper validation (CWE-502). Limiting factors that reduce real-world risk: (1) the attack vector is local (AV:L), not network-reachable, so the attacker must get their malicious file/page in front of the victim; (2) user interaction is mandatory (UI:R) - the victim must actively visit a malicious page or open a malicious file; (3) exploitation targets an environment that already has HunyuanDiT installed and is willing to ingest external model/merge files. There is no authentication barrier (PR:N), and successful exploitation runs with root privileges, so environments that routinely load third-party model artifacts are most exposed.

This is a genuine but conditionally-exploitable code-execution flaw rather than a drop-everything emergency. The provided CVSS 3.0 vector (AV:L/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H, base 7.8 HIGH) is internally consistent with the description: exploitation is local (AV:L) and requires user interaction (UI:R) - the victim must visit a malicious page or open a malicious file - but needs no prior authentication (PR:N) and yields full confidentiality, integrity, and availability impact because code runs in the context of root. The signals mostly align and point AWAY from mass exploitation: EPSS is only 0.51% (42nd percentile), the CVE is NOT in CISA KEV, and no public exploit was supplied in the input. The deserialization pattern (CWE-502) is a historically reliable RCE class, so weaponization is plausible if a researcher publishes a gadget chain, but as of this analysis no public exploit is identified. The key tension to flag: the AV:L + UI:R combination means this is NOT remotely wormable - it depends on luring a user who runs the HunyuanDiT app into processing an attacker-controlled model/merge artifact, which caps real-world blast radius. Net priority: moderate - patch promptly in any environment running HunyuanDiT that ingests untrusted model files, but this does not carry the urgency of a network-facing, no-interaction RCE.

An attacker crafts a malicious serialized model/merge artifact and distributes it (e.g., on a model-sharing site or via a malicious page/file) to a user running Tencent HunyuanDiT. When the victim loads or merges that artifact through the vulnerable merge endpoint, the untrusted data is deserialized and attacker-controlled code executes as root. No public exploit was provided in the input, but the CWE-502 deserialization pattern and local attack vector (AV:L/UI:R) make a working gadget chain realistic for a motivated researcher.