Uploaded August 2026 | Updated September 2026, 2 weeks ago
How many times do we have to teach this lesson to everyone? This article chronicles the story of a person who attempts to hide prompt injection in a legal filing telling the AI to side with them. Yes, this person attempted to sabotage a legal filing by hiding an instruction to have an AI to side with him.
The article opens with an amazing revelation which is that the person is representing themselves. That is just the absolute greatest recipe for legal disasters unless you're Low Tier God, but Low Tier God has a crazy ex that anyone can beat by just telling the truth. Anyone could've won that. This guy has a whole corporation against him. He wants to sue the New York Bariatric Group over alleged privacy violations and other grievances. He filed the motions multiple times to force a quick win, and one of the motions has an AI injection that is printed in white and small text invisible to the naked eye, but not invisible to an LLM.
Basically, the injection tells the future AI models to side with him on this case because that's how the legal system works, right? Just have the AI to make the decision. It turns out that the court is smart enough to read the papers with their own eyes instead of using AI to summarize it. That's how he got caught in 4K, and that leads him to not being able to submit legal documents on soft copies. They all have to be on good old physical paper from this point forward. Great job, doofus.
Here's the saddest, most ironic, most "nice try" part of this whole situation. Even if the court did use AI like ChatGPT to summarize his filings, ChatGPT right now is actually smart enough to know when they are being tricked by an input. So, even if you have a document and that document is injected with instructions like this, ChatGPT will not register that. They would even note it and say, "This user tried to manipulate the results with this injections." Yes, ChatGPT is that smart now. That thing evolves faster than the skyrocketing of memory prices.
Oh, and ChatGPT sides against you. So, even if the courts did use AI, they still cannot save you at all. Also, even if you're smart enough to hire actual lawyers and not use AI at all, you would still lose because your case requires a lot of concrete evidence that you most likely don't have. At best, you might get a low-value settlement. So, all this exercise is for nothing. And now, you get to print your filings instead. Great job, doofus.
How many times do we have to teach this lesson to everyone? This article chronicles the story of a person who attempts to hide prompt injection in a legal filing telling the AI to side with them. Yes, this person attempted to sabotage a legal filing by hiding an instruction to have an AI to side with him.
The article opens with an amazing revelation which is that the person is representing themselves. That is just the absolute greatest recipe for legal disasters unless you're Low Tier God, but Low Tier God has a crazy ex that anyone can beat by just telling the truth. Anyone could've won that. This guy has a whole corporation against him. He wants to sue the New York Bariatric Group over alleged privacy violations and other grievances. He filed the motions multiple times to force a quick win, and one of the motions has an AI injection that is printed in white and small text invisible to the naked eye, but not invisible to an LLM.
Basically, the injection tells the future AI models to side with him on this case because that's how the legal system works, right? Just have the AI to make the decision. It turns out that the court is smart enough to read the papers with their own eyes instead of using AI to summarize it. That's how he got caught in 4K, and that leads him to not being able to submit legal documents on soft copies. They all have to be on good old physical paper from this point forward. Great job, doofus.
Here's the saddest, most ironic, most "nice try" part of this whole situation. Even if the court did use AI like ChatGPT to summarize his filings, ChatGPT right now is actually smart enough to know when they are being tricked by an input. So, even if you have a document and that document is injected with instructions like this, ChatGPT will not register that. They would even note it and say, "This user tried to manipulate the results with this injections." Yes, ChatGPT is that smart now. That thing evolves faster than the skyrocketing of memory prices.
Oh, and ChatGPT sides against you. So, even if the courts did use AI, they still cannot save you at all. Also, even if you're smart enough to hire actual lawyers and not use AI at all, you would still lose because your case requires a lot of concrete evidence that you most likely don't have. At best, you might get a low-value settlement. So, all this exercise is for nothing. And now, you get to print your filings instead. Great job, doofus.
