OmniGen2-RL RCE (CVE-2026-25873)
CVE-2026-25873
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 req...
Overview
A critical security vulnerability has been discovered in OmniGen2-RL, specifically within its reward server component. Tracked as CVE-2026-25873, this flaw allows a remote attacker with no credentials to execute arbitrary commands on the host system. The vulnerability stems from the insecure deserialization of data in HTTP POST requests, providing a direct path for attackers to take full control of affected servers.
Vulnerability Details
In simple terms, the reward server in OmniGen2-RL improperly processes incoming data. When it receives certain HTTP POST requests, it uses a Python process called “pickle deserialization” to interpret the data. This process is inherently risky if applied to untrusted input, as it can be tricked into executing malicious code embedded within the request. Because the service does not require any authentication, any attacker who can reach the server’s network port can send a specially crafted request to exploit this weakness.
Impact and Risk
The impact of this vulnerability is severe. Successful exploitation grants an attacker the ability to run any command on the underlying operating system with the same privileges as the OmniGen2-RL service. This could lead to:
- Complete compromise of the server.
- Theft or destruction of sensitive data and models.
- Installation of persistent malware or ransomware.
- Use of the server as a foothold to attack other internal network systems.
Given the high ease of exploitation and the lack of required authentication, this vulnerability has received a Critical CVSS score of 9.8. Systems exposed to the internet are at immediate risk. For context on how such vulnerabilities lead to real-world incidents, you can review historical data breach reports at breach reports.
Remediation and Mitigation
Immediate action is required to protect your systems.
Primary Action: Patch or Update Contact the vendor of OmniGen2-RL immediately to obtain a patched version of the software. Apply the update to all affected instances as soon as possible. There is no safe workaround for this fundamental flaw; patching is the only complete solution.
Temporary Mitigation (If Patching is Delayed):
- Network Isolation: Ensure the reward server component is not accessible from the internet. Restrict network access to it using firewalls, allowing connections only from strictly necessary, trusted IP addresses.
- Monitor for Exploitation: Review server logs for unusual HTTP POST activity to the reward server endpoint. Implement intrusion detection rules for signs of pickle deserialization attacks.
Stay informed on emerging threats and patches by following the latest security news. After applying the patch, monitor your systems closely for any signs of prior compromise.
Never miss a critical vulnerability
Get real-time security alerts delivered to your preferred platform.
Related Advisories
PraisonAI is a multi-agent teams system. Prior to 4.5.115, the AgentService.loadAgentFromFile method uses the js-yaml library to parse YAML files without disabling dangerous tags (such as !!js/functio...
Deserialization of Untrusted Data vulnerability in Apache Storm. Versions Affected: before 2.8.6. Description: When processing topology credentials submitted via the Nimbus Thrift API, Storm deseri...
In Infoblox NIOS through 9.0.7, insecure deserialization can result in remote code execution....
Pachno 1.0.6 contains a deserialization vulnerability that allows unauthenticated attackers to execute arbitrary code by injecting malicious serialized objects into cache files. Attackers can write PH...