Summarize the opposing side's filings
Point Hexagone AI at the case file. Ask Claude for a summary of the weak arguments. Real identities stay with you.
Reading an opponent's filings closely is exactly the work an AI is good at, and exactly the work you cannot outsource. A litigation file identifies the parties on nearly every page. Redacting it by hand costs more time than the summary saves, and blacking out names leaves the facts, which in a small jurisdiction identify the case just as reliably.
How it works
Protect the whole file at once
Point Hexagone AI at the case folder. Scanned exhibits are read by local OCR, so photocopied filings are covered like the rest.
Ask the analytical question
Weak points, contradictions between exhibits, arguments left unanswered. The reasoning does not need to know who the parties are, only how the case hangs together.
Read it back in clear
Because the same party carries the same placeholder throughout, the summary reads as a coherent account of the dispute once the markers resolve on your machine.
Summarize the weak arguments in the opposing filings in [c.DUPONT_MARTIN].
What gets masked here
- Party and witness names
- Case and docket numbers
- Addresses and places of residence
- Dates of birth and identity numbers
- Amounts claimed and bank details
What you get back
A structured read of the opposing argument in minutes rather than an evening, with the residual re-identification risk measured rather than assumed.
How accurate is the detection?Questions on this use case
- Is anonymizing a case file enough to satisfy professional secrecy?
- The test is whether re-identification is reasonably likely, not whether names were removed. That is why Hexagone AI is evaluated on re-identification risk rather than recall: on the RAT-Bench test set from Imperial College London it records the lowest rate of the tools compared, 14% at level 1 and 27% at level 2, with no direct identifier recovered. With the desktop app the question is narrower still, because the original file never leaves your machine at all.
- What about exhibits that are scans rather than digital text?
- They are handled. Local OCR reads image-only PDFs before detection runs, which matters because a tool that silently skips scans gives you a false sense of coverage on exactly the documents most likely to be photocopied.
Try it on this exact task.
One week free, no credit card. Nothing is uploaded, so you can run it on a real file without asking anyone's permission.
macOS Apple Silicon and Windows 10/11