CVE-2026-115800: UbiquitousLearning mllm Convertor __init__.py load_model path traversal
A vulnerability was identified in UbiquitousLearning mllm up to 2.0.2. The affected element is the function loadmodel of the file pymllm/mobile/convertor/init.py of the component Convertor. The manipulation of the argument shardfile leads to path traversal. The attack needs to be performed locally. The exploit is publicly available and might be used. The project was informed of the problem early through an issue report but has not responded yet.
Affected Software
Event History
Frequently Asked Questions
Who is realistically exposed to this issue?
Systems using UbiquitousLearning mllm versions up to 2.0.2 and invoking the Convertor component's load_model function are affected. Exploitation requires local access and privileges sufficient to supply or influence the shard_file argument.
What does an attacker need to exploit it?
An attacker needs local access and the ability to manipulate the shard_file argument passed to load_model in pymllm/mobile/convertor/__init__.py. Public exploit code is available, which may make exploitation more practical where that input can be controlled.
Is a fix available from the project?
The provided information does not identify a fixed version. The project was notified through an issue report but had not responded at the time of publication.
What can be done while no patch is available?
Restrict who can execute local workflows that call load_model and prevent untrusted users or inputs from controlling shard_file. Validate and constrain shard_file values so they cannot reference paths outside the intended model-shard directory.