| Title | lm-sys FastChat ≤ 0.2.36（“up to 0.2.36”） Deserialization of Untrusted Data;Improper Input Validation |
|---|
| Description | A deserialization vulnerability exists in fastchat/model/apply_delta.py (functions split_files / apply_delta_low_cpu_mem) in FastChat ≤ 0.2.36. In our controlled tests, exploitation led to modification of model inference behavior (altered logits), and in multi-process environments persisted weight files were modified such that other processes observed integrity changes (Scope: Changed). The reported observations were performed in a controlled environment and are documented in a private demonstration video. No PoC is published publicly; encrypted evidence can be provided to CNA/NVD on request. |
|---|
| Source | ⚠️ Advisories: nvd.nist.gov/vuln/detail/CVE-2025-3677; github.com/lm-sys/FastChat/issues/3713. Exploit: no public PoC; private video on request. |
|---|
| User | Tianrui Song (UID 92554) |
|---|
| Submission | 11/12/2025 07:22 (11 months ago) |
|---|
| Moderation | 11/16/2025 17:04 (4 days later) |
|---|
| Status | Duplicate |
|---|
| VulDB entry | 304966 [lm-sys fastchat up to 0.2.36 apply_delta.py split_files/apply_delta_low_cpu_mem deserialization] |
|---|
| Points | 0 |
|---|