> but the facts that he was in possession of were not only widely known at the time but were the subject of an ongoing lawsuit that had launched months prior.
That is an exceedingly charitable read of these lawsuits.
Everyone knows LLMs are copyright infringement machines. Their architecture has no distinction between facts and expressions. For an LLM to be capable of learning and repeating facts, it must also be able to learn and repeat expressions. That is copyright infringement in action. And because these systems are used to directly replace the market for human-authored works they were trained on, it is also copyright infringement in spirit. There is no defending against the claim of copyright infringement on technical details. (C.f. Google Books, which was ruled fair use because of it's strict delineation of facts about books and the expressions of their contents, and provides the former but not a substitute for the latter.)
The legal defense AI companies put up is entirely predicated on "Well you can't prove that we did a copyright infringement on these specific works of yours!".
Which is nonsense, getting LLMs to regurgitate training data is easy. As easy at it is for them to output facts. Or rather, it was. AI companies maintain this claim of "you can't prove it" by aggressively filtering out any instances of problematic content whenever a claim surfaces. If you didn't collect extensive data before going public, the AI company quickly adds your works to it's copyright filter and proclaims in court that their LLMs do not "copy".
A copyright filter that scans all output for verbatim reproductions of training data sounds like a reasonable compromise solution, but it isn't. LLMs are paraphrasing machines, any such copyright filter will simply not work because the token sequence 2nd-most-probable to a copyrighted expression is a simple paraphrase of that copyrighted expression. Now, consider: LLMs treat facts and expressions as the same. Filtering impedes the LLM's ability to use and process facts. Strict and extensive filtering will lobotomize the system.
This leaves AI companies in a sensitive legal position. They are not playing fair in the courts. They are outright lying in the media. The wrong employees being called to testify will be ruineous. "We built an extensive system to obstruct discovery, here's the exact list of copyright infringement we hid". Even just knowing which coworkers worked on what systems (and should be called to testify) is dangerous information.
Sure. The information was public. But OpenAI denies it and gaslights extensively. They act like it's still private information, and to the courts, it currently still is.
And to clarify: No I'm not saying murder or any other foul play was involved here. Murder isn't the way companies silence their dangerous whistleblowers anyway. You don't need to hire a hitman when you can simply run someone out of town and harass them to the point of suicide with none of the legal culpability. Did that happen here? Who knows, phone & chat logs will show. Friends and family will almost certainly have known and would speak up if that is the case.