Vulnerability · Published Mar 18, 2026 · Assigned by VulnCheck
CVE-2026-25873OmniGen2-RL Reward Server Unsafe Deserialization RCE
Critical 9.8
In plain English
A critical-severity unsafe deserialization in OmniGen2-RL from Beijing Academy of Artificial Intelligence (BAAI). Proof-of-concept exploit code is public. The chance it is exploited in the next 30 days is estimated at 1.1% (higher than 67% of all known vulnerabilities).
Official description
OmniGen2-RL contains an unauthenticated remote code execution vulnerability in the reward server component that allows remote attackers to execute arbitrary commands by sending malicious HTTP POST requests. Attackers can exploit insecure pickle deserialization of request bodies to achieve code execution on the host system running the exposed service.
Affected products (as reported)
| Vendor | Product | Versions |
|---|---|---|
| Beijing Academy of Artificial Intelligence (BAAI) | OmniGen2-RL | 0 |
Weakness: CWE-502 · Vector: CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H
References
- https://github.com/VectorSpaceLab/OmniGen2/pull/139 issue-tracking
- https://chocapikk.com/posts/2026/omnigen2-pickle-rce/ technical-description, exploit
- https://arxiv.org/abs/2506.18871 product
- https://github.com/VectorSpaceLab/OmniGen2/blob/3a13017e532f9f309a38bca571fd62200a6415c5/OmniGen2-RL/reward_server/reward_server.py#L118 related
- https://github.com/VectorSpaceLab/OmniGen2/blob/3a13017e532f9f309a38bca571fd62200a6415c5/OmniGen2-RL/reward_server/reward_proxy.py#L208 related
- https://github.com/VectorSpaceLab/OmniGen2/blob/3a13017e532f9f309a38bca571fd62200a6415c5/OmniGen2-RL/reward_server/reward_proxy.py#L224 related
- https://www.vulncheck.com/advisories/omnigen2-rl-reward-server-unsafe-deserialization-rce third-party-advisory