Post

Free Law Project ⚖
@free.law

Ok, fine, we’ll add this as a feature in our PDF analysis tool: github.com/freelawproje...

github.comFind prompt injection · Issue #227 · freelawproject/x-rayI am not convinced that this is the same category of problem as bad reactions, per se, but we'll be able to flag it the same as everything else, so we might as well. It'll be a nice feature to add ...
Techdirt@techdirt.com

Pro Se Plaintiff Caught Hiding Prompt Injections In Court Filings; Responds By Hiding More There have now been dozens of stories of bad lawyers using AI results in... https://www.techdirt.com/2026/08/14/pro-se-plaintiff-caught-hiding-prompt-injections-in-court-filings-responds-by-hiding-more/

techdirt.comPro Se Plaintiff Caught Hiding Prompt Injections In Court Filings; Responds By Hiding MoreThere have now been dozens of stories of bad lawyers using AI results in filings, most of which show up in the form of fake case citations. That's old hat at this point. But how about prompt injections in court filings? There have been a few people warning that it's coming. A year ago, Louisiana...

2 reposts18 likes

Reid Marlow
@reidmarlow.com

The hard part is treating the PDF as evidence and hostile input at the same time. Are you planning to isolate extracted instructions from document text, or just flag when the model starts treating filings as commands?

Reid Marlow
@reidmarlow.com

That catches the cheap class. I'd still want it reported as evidence metadata, not silently stripped, because courts will care whether it was a scanner artifact, a filing trick, or actual text. Different failure modes need different handling.