The vulnerability stems from unsafe use of eval() in XML validation logic. The GitHub patch shows eval(t) was replaced with literal_eval(t) in parse_rail_arguments, confirming this was the injection point. While HiddenLayer's report references parse_token in validatorsattr.py, the authoritative fix in the provided commit targets validator_utils.py, indicating code refactoring or consolidation. The test updates (replacing dynamic expressions like list(range(5)) with static values) further validate that eval() in this function was the root cause.