Blog
Benchmarks, methodology, engineering notes.
Legal AI vendor lock-in: who chooses your firm's model?
Harvey put its own model into its product this summer. What happened, why it matters for control, and four things a firm can keep in its own hands.
Read the postOpinionThe end of prestige
A model now hands anyone a well-structured message in seconds. The price of sounding credible went to zero, and the proxy went with it.
Read the postFuture of WorkTowards the age of AI disclosure
Most people hide their AI use. There is no policy, peers judge it, and it makes work look cheap. Rivals who disclose and charge less will end the secrecy.
Read the postLegalYour small firm's AI policy leaves no record
Clio finds 75% of small practices use AI and 55% with no policy. The North Carolina Bar finds 9% enforce one. Rules are cheap; proof is not.
Read the postEngineeringHow we shrank our AI model to fit a browser tab
Our detection model carried a vocabulary for 104 languages. We used seven. How we cut 39 MB without changing the model, and why a browser tab is the hard part.
Read the postFuture of WorkWe built the perfect conditions for hiding AI use
Admit you use AI and your colleagues call it cheating; be overheard and your employer bans it. Hiding is rational, and it makes AI policy unenforceable.
Read the postLegalRedact legal PDFs without the cloud or fake black boxes
A black box drawn in a PDF editor can leave the text underneath. How to redact client PDFs on your own machine, step by step, and how the three options compare.
Read the postHow it worksReversible redaction: how tokenisation lets you reverse it
Reversible redaction swaps each name for a code and keeps the map on your machine. Share the document, decode it later. Codes are per folder, not global.
Read the postComplianceGDPR anonymization: what actually counts as anonymised
The test the GDPR applies, why keeping a mapping table changes the answer, and a quick guide to choosing anonymization or pseudonymization.
Read the postGuidesHow to redact scanned PDFs when your tool finds nothing
If your redaction tool found no personal data in a scan, the scan is why. The three ways redacting a scan fails quietly, and how to do it on your own device.
Read the postAI SafetyIs uploading a contract to ChatGPT safe? The clause to check
Switching off training does not settle it. The confidentiality clause most people skip, what a contract really exposes, and what to do if you already uploaded.
Read the postEngineeringWhy we built our own demo studio
We started with an off-the-shelf demo-video tool. Two themes, seven languages and an interface that changes every week broke it, so we built our own.
Read the postLegal AIAre AI chats privileged? The three rulings so far
Three US courts ruled on AI chats and privilege in early 2026 and did not agree. What decided each case, and what to tell clients who use chatbots on their own.
Read the postLegal AICopilot for Legal, Claude for Legal: where do documents go?
Four AI companies launched legal products in 2026. Inside a single one, the answer to a client's data question can change with an admin setting. What to check.
Read the postDevelopersA redaction API that runs inside your own perimeter
Every cloud redaction API asks you to upload the sensitive documents first. Ours runs on your own host, as a local REST service or a command-line tool.
Read the postLegal AIWhy pay for Harvey if it runs on Claude and GPT?
Harvey uses models any firm can subscribe to. What the premium buys instead, and whether a firm can get it without a 100-seat, one-year contract.
Read the postPositioningWhy nobody built a redaction tool for you
Almost every PII tool takes text, not a page. It starts after the hard part is done and never says whose job that part is.
Read the postWorkflowConsistent codes across documents: where they hold
A name that changes code every turn makes an AI conversation unreadable. How far a code stays the same, what changes it, and where the table is kept.
Read the postEngineeringWhy we built our own PII engine instead of shipping Presidio
Presidio is an excellent toolkit. We still wrote our own engine, because our users work on a single PDF on a laptop, not a text stream in a cloud pipeline.
Read the postPositioningHow companies protect PII, and the gap we fill
Companies protect personal data in four ways: governance platforms, cloud APIs, managed redaction and manual blackout. A plain map, and the gap we fill.
Read the postResponsible AIHuman-in-the-loop anonymization: who signs off
Signing what you have not read puts your name at stake. What an anonymization tool must let you check before it produces the file.
Read the postAI SafetyIs it safe to paste documents into ChatGPT? 10-second check
Your account tier decides less than you think. What happens to pasted text, what the Samsung leak settled, and the ten-second check to run before you paste.
Read the postPrivacy EngineeringPII protection is different when AI is in the loop
For 30 years, protecting personal data meant controlling database access. AI created an easy path from controlled documents to the public cloud.
Read the postData RiskIf you quietly share client documents with a chatbot
AI makes exposing confidential data effortless, and the consequences are real. What you risk when you paste client files into a chatbot.
Read the postBenchmarksPresidio vs. promptShield benchmark: what each tool missed
We labeled every sensitive word in 48 fictional documents in eight languages. What each tool left readable, where it missed, and how to rerun the test.
Read the post