A security researcher found alarming function names and prompt-injection strings in a website fingerprinting script. The analyzed behavior collected browser characteristics such as canvas and WebGL signals, but some AI systems refused to analyze the code after reacting to its wording while another model continued. The case illustrates the weakness of inferring execution semantics from names and strings alone. The author’s intent remains an inference, the observed reactions do not generalize to every AI system, and the text is not evidence that the harmful capabilities it describes were implemented.