safety properties of direct execution of ai-generated code

[fche]

This line seems to straight execute python code that comes from the LLM.

https://github.com/gd3kr/BlenderGPT/blob/990068512ed9ee0c37091a3295157724e5a0dd13/init.py#L254

What if any assurance exists that the code cannot possibly contain malware that could damage the blender internal data, or perhaps even run arbitrary code with the invoking user's privileges?