The verification layer

Verify any AI output. From any tool.

Your firm already has AI in the building. Harvey, Copilot, ChatGPT, ours, or all four. Send any output here and get back an independent verdict: every citation checked against the authority itself, the SRA-mapped policies evaluated, and a named lawyer's sign-off recorded as evidence you can produce later.

No integration needed on day one. Paste, upload, email in, Word pane, or API.

app.legalaispace.com/check

Check this

Submit any AI output for independent verification. The result and your sign-off become signed evidence in the firm's record.

Paste text, or drop a document

.docx · .pdf · plain text · or email it in

Produced byHarveyMicrosoft CopilotChatGPTCoCounselOur agentsOther
Matter: illustrative matter referenceRun check

What runs: citation extraction, verification against the allowed sources, good-law treatment, the SRA-mapped policy evaluations, then a readiness verdict.

Why it has to be independent

“Do not trust an AI system to judge its own accuracy, remembering that current AI does not have a concept of truth.”

SRA Risk Outlook report, The use of artificial intelligence in the legal market, 20 November 2023.

Every AI vendor now ships its own checking. That is worth having, and it is not what the regulator described. A system assessing its own output is the audited choosing the auditor. It also sees one tool, while a firm runs several. Verification that is independent of the model being checked, and that looks across the whole estate rather than one vendor's corner of it, is a different thing entirely.

What happens

Six steps, then a record.

The same sequence runs whether the output came from your AI assistant, a general-purpose chatbot, or one of our agents.

01

Submit

Paste it, upload it, email it in, or use the Word pane. Tag the tool that produced it and the matter it belongs to.

02

Extract

References and claims are parsed out of the text into a citation ledger, automatically.

03

Verify

Each citation is fetched from the authority itself and the quoted words are matched against the source.

04

Evaluate

The SRA-mapped policies run as code. Failures that should block a deliverable are marked as blocking.

05

Review

A plain verdict with specific action items. Never a confidence score.

06

Sign off

A named lawyer accepts responsibility. The signed evidence record is minted at that moment.

The verdict

Black and white, never grey.

No confidence percentages. Three states, and every item that is not green tells you exactly what to do about it.

app.legalaispace.com/check

Solid. Confirm two points before relying on this output.

9 of 11 citations verified · 11 of 12 policies passed

Limitation Act 1980, s.32

Verified

None

Civil Evidence Act 1968, s.11

Verified

None

Authority C (illustrative)

Needs a check

Read the paragraph. The output overstates the holding.

Authority D (illustrative)

Not found

Treat as fabricated. Remove before use.

Plus seven more verified references. Policy results: confidentiality passed, information barriers passed, no-hallucinated-citations failed and is blocking, audit trail complete.

Sign off as reviewedExport evidence recordSign-off is recorded as a signed audit event naming the reviewer.

Illustrative. Passing rows name real authorities because “this exists and the quote matches” is a true statement about them. Failing rows use placeholders, because printing a real citation beside the word fabricated would assert something false about a real judgment.

What the fee-earner gets

  • A defensible answer in minutes, rather than a review chore you do twice.
  • Re-check any single citation on demand, without rerunning the whole thing.
  • The invented case is caught before it reaches a filing, not after.

What the firm gets

  • Evidence of supervision on every AI-touched output, not just the ones someone remembered to check.
  • Fabrications and policy failures land on the COLP worklist with an action attached.
  • Per-tool quality statistics, so you learn which AI can be trusted for what.

Where it runs

The platform cannot demand a new habit.

Verification goes to the document, the inbox and the firm's own systems, rather than asking a fee-earner to remember a new tab.

The Word pane

Verification comes to the document. Flagged passages highlight in place, and sign-off happens without leaving Word.

Email it in

Forward any output to your firm's check address. The verdict and evidence link come back to your inbox. Zero training required.

The API

For firms that want their existing tools wired straight in, or a channel partner building on top.

Questions

Verifying AI output, answered.

The questions COLPs and fee-earners ask before they put an output through anything.

Still have questions? Talk to the founder
Because the SRA's own guidance says you should not. Its Risk Outlook report on artificial intelligence in the legal market, published 20 November 2023, tells firms: "Do not trust an AI system to judge its own accuracy, remembering that current AI does not have a concept of truth." A tool grading its own homework does not answer that. An independent layer does, and it is the same reason an auditor is not chosen by the audited.

From the blog

Why verification is the part that matters

What goes wrong when AI output reaches a filing unchecked, and the method for catching it.

Put one of your own AI outputs through it.

Bring something a tool in your firm produced this week. We will run it and show you what comes back.

Start the readiness check