Dshamir stops treating AI-Legal's backend like a prototype
A batch of work turns the server from "trust me" into something that checks what it's handed and keeps a clean record of what it does.
The most tangible change: uploaded files are now checked for what they actually are, not just what their name claims. Rename a spreadsheet to end in ".pdf" and the old setup might have waved it through; now the server inspects the file's real fingerprint and only lets genuine PDFs and Word documents past. Everything else gets turned away.
The rest is quieter but matters for anyone trusting this tool with confidential matters. Logs now automatically scrub out passwords, keys, and auth tokens before they're written down, so credentials don't leak into operational records. Errors get captured and tagged so a problem can be traced end to end, and the browser-facing security rules - previously switched off - are now actually enforced.
Spotted something wrong? Or know the PR text has fresher detail than the writeup above?