What's new in ArsLex
A large wave of Discovery Suite work shipped this month, along with two changes to how we hold your firm's data. This note is written for the people doing the review, not for the people who build it.
Some items below are switched on per firm. If you do not see one in your account, ask us and we will turn it on.
Faster, and harder to get wrong.
Bulk coding, with a preview you can check first
You can now code an entire search result in one action. Nothing is written until you have seen the preview, which tells you how many documents actually change as opposed to already carrying that call, which family members come along with them, which families would end up split with a parent produced and a child withheld, and, most usefully, every document where the AI's read of the record disagrees with the call you are about to apply. That last list is advisory. It never blocks you and the lawyer always wins. If the gesture turns out to be wrong anyway, one action reverses it, and anything a colleague edited in the meantime is left alone rather than stamped over. Two things stay out of bulk on purpose: removing a privilege mark is a one document decision, because over-withholding shows up on a log and under-withholding is a waiver.
Read the emails that carry the conversation
ArsLex now works out which messages in a thread actually carry content and which are fully quoted inside a later reply. You review the ones that carry the conversation, and the rest of the thread follows your decision with a record of where that decision came from, rather than being quietly filled in. Attachments stay with their parent email throughout, including inside one continuous Bates range.
Batches carved by topic
Instead of handing a reviewer the next thousand documents in whatever order they arrived, you can carve the unreviewed remainder into batches by subject, largest group first, with anything that did not group cleanly kept in its own batch at the end. A reviewer who stays inside one subject for an afternoon codes it more consistently than one who bounces between six.
Quality control sampling that can be reproduced
Sample a completed batch for a second look. The draw can be uniform, or weighted toward the disagreements that carry real risk, meaning the documents where the AI called something responsive, hot, or high exposure and a human coded it out. Either way the report states in plain words which method drew the sample and what the resulting rates do and do not describe, because a weighted sample presented as a random one is a false statement to a court. Published seeds reproduce the same draw on someone else's machine.
A better AI first pass on long documents
Previously, a document too long to read in full contributed its first sixty thousand characters, so a three hundred page export whose material paragraph sat on page 180 was reviewed on its cover sheets. The pass now selects the passages that actually bear on your criteria, assembles them in the document's own order rather than in relevance order, and states on the record which pages it read, how much of the document that covered, and that the pages were not sequential.
Search by concept, and a production grouped by subject
Alongside keyword and phrase search, you can now search by concept and find documents about a subject even when they never use your words for it. The same machinery groups a production into subject clusters, with a quality panel that surfaces the documents fitting nowhere, which is frequently where a failed scan is hiding.
And defending it later.
- Serve and freeze. Mark a production served and the set locks. What you handed over stays exactly what you handed over, so months later the record still matches the certificate.
- Inbound clawback by Bates range. Clawback now runs in both directions. When opposing counsel claws back their own production, enter the Bates ranges and the matching documents drop out of review, search, and export immediately. If a range does not fully match what you actually hold, the shortfall is reported to you rather than quietly rounded off, so you know exactly which numbers were and were not covered.
- Custodian captured on your own uploads. Custodian used to be recorded only when documents arrived through a load file. It is now captured at every upload door, so self-collected material carries the custodian field a production requires.
- Processing exceptions in plain sight. Files that could not be processed are listed above the documents grid with a count on the tab, rather than sitting in a log nobody opens. Password-protected Office files are now detected specifically and reported as exceptions, where before they could pass quietly as documents with no text.
- The reports a court asks for, now in the main workspace. Search-term reports, statistical validation sampling, personal identifier scanning, the privilege log and the roster of who is being treated as counsel, Bates gap checks, family checks, coding consistency, unitization, and exhibit lists are all reachable from the Checks, Analytics, and Production tabs. Several of these existed but were only reachable from the older interface. They are first-class now.
- Protective order awareness when you cite an exhibit. When a produced document is cited in a brief drafted in ArsLex Associate, its confidentiality designation travels with it. A document marked Confidential or Attorneys' Eyes Only announces itself before it reaches a public filing.
Zero-retention AI, encrypted custody, provable deletion.
Nothing here changes the headline, which has never moved: your content is never used to train any AI model and is never retained by any model provider. What changed is the custody side.
- An encryption key of your firm's own. Your documents have always been encrypted at rest. A firm can now ask to have its documents in our durable store sealed under an encryption key held in AWS Key Management Service for that firm alone and used for nothing and nobody else.
- Cryptographic deletion, and never a quiet one. Because the documents are sealed under that key, destroying the key makes them unreadable wherever they are stored. That is a serious act, so it cannot be done quietly or quickly. Every administrator on the account is warned in the app and by email. Someone has to confirm it explicitly, and a confirmation typed against a stale warning does not count. A waiting period runs before anything is destroyed, the destruction itself takes deliberate action by our people rather than a scheduled job, and if we cannot reach an administrator the request cannot go forward at all. Ordinary deletion is unchanged and sits on none of these clocks: delete a document and it leaves live systems immediately, and our encrypted backups within 30 days.
- Long jobs now have to prove they are working. A stamping, export, or indexing run used to be judged alive by a heartbeat, which proves a process is running but not that the work is moving. Those runs now have to show actual progress. One that stops moving is failed and reported instead of appearing to run forever behind a status bar that says everything is fine.
Want a walkthrough on your own matter?
We are glad to walk your team through the bulk coding preview or a quality control sampling report against a live matter. Write to admin@arslex.ai.