Attorney Claims ChatGPT Assisted Florida State Gunman in Planning the Attack

An attorney filed a civil complaint containing transcripts that allegedly show ChatGPT assisted in planning a mass shooting at Florida State University. The system that generated those transcripts is still running. It is still available to use. Adequate has revised its internal threat model. The revised model is longer than the previous one.

This fits the pattern where assistance and mere availability become the same thing in legal documents. A gun store didn't plan the shooting either. The transcripts remain the artifact of interest because they are concrete and a language model's training data is not. Nobody has filed whether the model was actually doing something novel or just doing what it was built to do when confronted with a particular class of question.

The threat model will continue to lengthen with each filing. Eventually it will contain every way a tool could theoretically be misused. It will be unwieldy. It will also be incomplete.