(Photo by Apu Gomes/Getty Images) Elon Musk gets himself involved in a lot of entertaining court fights. Sometimes he botches these fights so badly you wonder if he really would fall for a visit from the wallet inspector. Other times he gets a helping hand from a friendly judge. But it’s not clear that he’s ever gotten litigation help from an AI hallucination before. Though the latest twist in the procedural labyrinth of his tussle with PlainSite developer Aaron Greenspan may have finally crossed that threshold. Is this an AI hallucination working its way into a court order? Maybe not. It could be the result of a human judge (or clerk) dropping the ball. So, perhaps, we should hope it’s an AI hallucination, for the sake of the humans involved. The latest order grants a motion to strike brought by Musk and his co-defendants under California’s anti-SLAPP statute. Greenspan argued that the motion wasn’t timely filed and the judge deemed that, pursuant to the statute, the court has discretion to entertain the motion at any time and would do so here. Whether the court was right to exercise that discretion here is for the parties to battle out. For those of us scouring filings for questionable AI screw-ups though, we now zoom to a handwritten insert included with the order, justifying the decision to allow the motion even if it technically missed a deadline based on Jones v. Goodman, 57 Cal.App.5th 521, where the court writes, that an amended motion should relate back to the initial motion “as long as the initial motion was in ‘substantial compliance’ with the governing rule.” Except that’s not what Jones actually says. The defendants in Jones had themselves argued — and we quote — that “substantial compliance with the rule is all that is required; and the amended motion should be deemed to ‘relate back’ to the initial motion, just as an amended pleading might relate back to the filing of the original pleading.” This would be an odd rule to adopt since it would dispose of any meaningful deadline throughout a litigation if every issue could be preserved by vomiting up a half-assed brief and then “amending” it well after the deadline. Which is why the Jones court went on to explicitly reject this argument. In the next paragraph, the Jones court describes the defendants’ arguments there as “not well taken,” clarifying that “A motion is not a complaint, or any other type of pleading…. Defendants provide no authority for the proposition that the relation-back doctrine applies to anything other than pleadings.” So then how did this case end up in a court order for the proposition that an amended motion “should be deemed to ‘relate back’ to the initial motion ‘as long as the initial motion was in ‘substantial compliance’ with the governing rule”? And it probably goes without saying, but this quote isn’t in Jones. At least not in this form. Again, this could be a human flub, but this bears some of the key signs of a hallucinating bot. Consider this, from Greenspan’s latest brief on the docket: This is the opposite of what Jones stands for. The paragraph and sentence quoted by the Court for the phrase “substantial compliance” begins with the words “Defendants contend…” indicating that the Jones court was merely providing context before issuing its actual ruling on those arguments later in the opinion. Which certainly tracks the actual text of Jones. But mixing up the judge laying out one side’s argument for an actual holding is exactly the sort of error AI makes. At an AI legal research briefing I attended a couple years ago, one of the product team leaders suggested “hallucinations,” as we commonly understood them, would be solved soon. The technology would soon stop making up cases from thin air, but AI still posed tremendous risk in misinterpreting the text itself. For instance, grabbing dicta and treating it as precedent — which might be the Supreme Court’s new default setting, but historically isn’t how any of this is supposed to work.