AI you can trust defend,
wired into the work
lawyers actually do.

graphlion reads the whole evidence base behind a subject access request, a disclosure request or an investigation, shows exactly how it reached every answer, and carries the matter from first question to a defensible answer.

Get a demo See how it proves itself

Getting an answer is the easy part now.

Any model can read a document and tell you something plausible. None of them can tell you what they missed, why they decided what they decided, or hand you something that holds up when opposing counsel, a regulator or a judge pulls at it. That gap is where the risk sits, and it is the whole reason graphlion exists.

One matter, four stages

Not a research tool bolted onto a review platform. Asking, searching, reviewing and resolving the people involved happen in one place, on one record.

01

Ask

Put a question to the matter. graphlion plans the search, builds a working set and answers from it, with every source cited.

02

Searches

Working sets are saved, named and reusable. Each one keeps the record of exactly how it was built.

03

Review

Reviewers code batches for relevance and issues. The AI suggests, the lawyer decides, and every override is logged.

04

People and organisations

Aliases, misspellings and multiple addresses resolved to one person, with their whole chronology in one place.

Every classification comes with its own audit

graphlion never shows a confidence score without a route to the evidence behind it. Open any AI decision and you get the same four things.

Elusion test
A sample drawn from outside the set, and the misses found in it, reported as coverage over the corpus that was searched. It tells you what the search did not catch, not only what it did.
Confidence
Banded, never a bare number, and always paired with the documents that produced it.
Override history
Every human correction, and the reviewer-agreement rate that follows from them.
The exact prompt
The instruction that was applied, and its version. No black box to explain away.

All of it accumulates into a continuous methodology record: collection, search terms, de-duplication, AI seed, QC. When someone asks how the exercise was run, it is one export rather than a fortnight of reconstruction.

The obligation is to search well, and to show it

Organisations answering a subject access request, a disclosure request or an internal investigation carry a legal duty to run a search that is reasonable and proportionate, and to be able to justify it afterwards. The Data (Use and Access) Act put that standard on a statutory footing by inserting Article 15(1A) into UK GDPR, and from 19 June 2026 section 164A of the Data Protection Act 2018 requires every controller to handle complaints about it directly, before the ICO ever sees them. There is no technology exemption. A machine ran the search, and the duty is unchanged.

graphlion is built for that standard. Every answer arrives with a record of what was searched, what was found, what was not, and how confident the system is in the difference.

See it on your own matter

A short walkthrough on a real disclosure exercise, including the part where graphlion tells you what it might have missed.

Get a demo