Internagent
Monthly
InternScience InternAgent 1.5.0 contains an untrusted-control-sphere inclusion flaw (CWE-829) in the LoRA loading logic reached through the 'train' function in tasks/AutoChem/code/experiment.py, where an attacker-influenced base_model_name_or_path value causes functionality to be pulled from outside the intended trust boundary. The assessed impact is limited to integrity (CVSS 3.1: 5.3, AV:N/AC:L/PR:N/UI:R, I:L only, no confidentiality or availability impact), and exploitation is not passive - a legitimate operator must run the training routine with the malicious path value. A public PoC reference exists; it is not verified as working exploit code, and there is no KEV confirmation of active exploitation; this is a moderate-severity issue slated for this month's remediation cycle rather than an emergency.
InternScience InternAgent 1.5.0 contains an untrusted-control-sphere inclusion flaw (CWE-829) in the LoRA loading logic reached through the 'train' function in tasks/AutoChem/code/experiment.py, where an attacker-influenced base_model_name_or_path value causes functionality to be pulled from outside the intended trust boundary. The assessed impact is limited to integrity (CVSS 3.1: 5.3, AV:N/AC:L/PR:N/UI:R, I:L only, no confidentiality or availability impact), and exploitation is not passive - a legitimate operator must run the training routine with the malicious path value. A public PoC reference exists; it is not verified as working exploit code, and there is no KEV confirmation of active exploitation; this is a moderate-severity issue slated for this month's remediation cycle rather than an emergency.