College student went on a destructive rampage, then confessed to ChatGPT, cops say A Missouri college student has learned the hard way that admitting a vandalism spree to ChatGPT and asking whether he was likely to get caught may not be the best use of AI. According to court documents [PDF] filed in Greene County, Missouri circuit court, Missouri State University sophomore Ryan Schaefer is suspected of …
This wasn't one of them. The AI didn't, nor is an AI company claiming, detect the crime and notify the police. The police identified this beforehand and found written text on a phone they were handed even though the criminal didn't have to*. The AI did nothing at all other than write about the consequences of vandalism, probably incorrectly given it's a local crime and there are lots of different jurisdictions who may have different sentencing requirements and sublevels.
* This being the US, the police don't have the legal right to compel access to the contents of a device. Had the student locked the device with a code and declined to provide the code, they could have prevented having that log in the evidence pile, at least unless the police went to OpenAI with a warrant which they probably wouldn't need to bother doing. In some other countries including the UK and Canada, the police can demand access and punish you if they don't get it. From their statements, they asked for access and were given it, meaning the student was adding one more to a pile of stupidities. It probably wouldn't have made much of a difference if they had given the student had already been located.
Does it really count as an admission of actually having done the deed when you discuss it with ChatGPT?
I mean, if, entirely theoretically of course, and only done in order to play around with ideas for this short story I'm writing, which involves checking what sort of responses LLMs give when one pretends to it that something dreadful has been committed, purely for verisimilitude you understand, one happens to type something into ChatGPT late in the evening...
(Delete damn you, delete! Where is the AlkaSeltzer? Aaargh, blast you, stop that fizzing noise)
Anything you record/write/type into a computer or phone can be evidence, even if just circumstantial. If they'd found a diary with an entry from the day before saying "I rly feel lik fukinup some cars n brakin windzhaileds n shitz" that would be admissible. Not proof, but admissible evidence.
I imagine a even a half shonky lawyer† could make a case that his client's conversation with ChatGPT was covered by legal privilege and inadmissible.
† I understand Michael Avenatti should be "free" to take new cases from next year. Leaking Brains Rudy available now.
Everyone who's tried to use the services of AI, AI and AI, LLP has discovered that chatbots are not authorized to practice law. A lawyer trying to make the argument that chat history is protected by privilege would get benchslapped HARD and any other protections were waived when the idiot gave the police his phone.
