A critical Server-Side Request Forgery vulnerability in InternLM LMDeploy allows unauthenticated remote attackers to access sensitive internal metadata and network resources.

This vulnerability occurs in the load_image() function within lmdeploy/vl/utils.py, which fails to validate user-supplied URLs. An unauthenticated attacker can trigger this flaw to perform requests to internal or private IP addresses.

The vulnerability carries a CVSS score of 7.5, indicating a high level of risk due to the potential for unauthorized access to cloud metadata services and internal infrastructure. Successful exploitation may lead to the exposure of sensitive credentials, configuration data, or internal service information, which could facilitate further lateral movement within the network.

Immediate Action: Upgrade InternLM LMDeploy to version 0.12.3 or later to incorporate the necessary URL validation logic.

Proactive Monitoring: Review web server and application logs for suspicious outbound requests to internal IP ranges or cloud metadata endpoints (such as 169.254.169.254).

Compensating Controls: Implement egress filtering on the host machine to restrict outbound network access from the LMDeploy service to only essential external domains.

Public Exploit Available: No (The available data confirms the presence of a Nuclei detection template, but no weaponized exploit or public proof-of-concept exists).

The risk posed by this SSRF vulnerability is significant, as it provides a clear vector for internal reconnaissance and information disclosure. Organizations currently utilizing LMDeploy must prioritize the update to version 0.12.3 to neutralize this attack surface. Until an update can be applied, strict egress filtering should be enforced to prevent the application from communicating with internal network segments.