CVE-2026-46517 affects InternLM's lmdeploy toolkit for compressing, deploying, and serving large language models. In versions 0.12.3 and prior, the hardcoded configuration 'trust_remote_code=True' allows remote code execution through the Hugging Face supply chain without requiring user opt-in. This improper control of code generation (CWE-94) leads to a critical security risk. No official fix or patch has been released as of the publication date.

Successful exploitation of this vulnerability could allow an attacker to execute arbitrary code remotely with the privileges of the lmdeploy process. This compromises confidentiality, integrity, and availability of the affected system. The CVSS score of 7.8 reflects high impact on confidentiality, integrity, and availability. There are no known exploits in the wild at this time.

No official patch or remediation is currently available. Users should avoid using affected versions (<=0.12.3) of lmdeploy until a fix is released. Monitor the vendor's advisory channels for updates. Since the vulnerability arises from a hardcoded setting, disabling or overriding 'trust_remote_code=True' if possible may reduce risk, but this is not confirmed by the vendor.