We build AI workflows that hold up under attorney-client privilege, the professional code and GDPR. Pseudonymization, audit log, mandatory review — productive, not theoretical.
Three conflicts we see in every firm — and they're not solvable with off-the-shelf tools.
Staff paste briefs into ChatGPT for revision. Attorney-client material flows to a US provider with no professional confidentiality obligation — a clear §203 violation with damages and criminal-law exposure.
Clients increasingly expect faster turnaround, but every AI output must be lawyer-reviewed. Without clear workflows the time savings disappear into proof-reading — or worse, unreviewed AI outputs reach the client.
Many DPOs default to blanket AI bans because they lack the technical detail. Result: shadow IT — staff use ChatGPT anyway, just uncontrolled. We deliver an auditable architecture the DPO can actually approve.
Four use cases we've shipped most often in the last 18 months.
Inputs are pseudonymized before the model call (client name → 'Client_A'). The AI delivers a draft, re-identification happens locally. Lawyer reviews, finalizes, signs.
Effekt: 30–50% time saving on routine briefs, no client data ever reaches the model.
Connection to juris/beck-online/publisher databases. The AI structures search results, compares clauses, summarizes long judgments — without client context.
Effekt: Initial research outline in 5–10 minutes instead of 30–60.
Files are pseudonymized locally, sent to Azure OpenAI in EU region, the AI produces a structured summary referencing original document numbers.
Effekt: Onboarding new staff shortened by days.
AI generates draft responses for routine queries (scheduling, status updates). Lawyer sees the suggestion, releases or edits. Audit log documents every send.
Effekt: Response time on routine queries cut in half, secretarial workload reduced.
Clear separation: what off-the-shelf tools can do, where they fail, and where we step in architecturally.
| Bereich | Standard-ChatGPT | Was nicht geht | Unsere Architektur |
|---|---|---|---|
| Brief preparation | ChatGPT accepts client data but sends it to US servers. | OpenAI's standard DPA does not cover §203. Professional-code violation. | Local pseudonymization + Azure OpenAI (EU). Client data never leaves the firm in clear text. |
| Audit evidence | No logs, no versioning, no traceability. | No evidence preservation in case of dispute. | Full audit log: who processed what, when, on which model. |
| Client correspondence | No mandatory review, no release workflow. | Risk of sending unreviewed AI outputs. | Workflow forces lawyer release before send. Signature path documented. |
| Professional-code compliance | No confidentiality obligation on the AI provider. | No DPA in a §203-compatible setup. | DPA with a processor that commits to professional-code-equivalent measures. |
Sample calculation for a 6-lawyer firm with 4 secretarial staff.
Conservative estimate without research use case. Actuals depend on case mix — we run the numbers for your firm in the intro call.
Three clear stages — you join where it fits.
Workshop
Half day. Prioritize use cases, work through §203 risk, action plan.
Pilot
4–6 weeks. One use case pseudonymized, documented, in trial.
Roll-out
Expansion to more use cases, training all secretarial staff, DPA bundle.
Direkte Belege, dass Compliance-Software bei uns produktiv läuft – nicht nur in PowerPoint.
„Thanks to LiteLog we could prove we were at the right place at the right time, and that we did our job. Without that proof, we'd have paid 30,000 € in damages."
„The whole tool is well laid out and easy to use. Questions get answered quickly and without fuss. I can fully recommend it."
Standard ChatGPT with client data — yes. OpenAI is not a bound professional and data flows to the US. With pseudonymization before input plus Azure OpenAI in an EU region under a §203-compatible setup (confidentiality obligation of the processor, DPA with the specific operator), it becomes defensible. That's exactly what we set up.
Drafting briefs (with pseudonymized client details), case-law and literature research, clause comparisons, summaries of long files, client correspondence and onboarding forms. We recommend starting with two clearly scoped use cases — not a firm-wide all-in-one solution.
The lawyer — same as with any other tool. So we build workflows with mandatory review (a human checks every AI output before it touches the client), versioning and an audit log, so it's traceable in dispute what the AI suggested vs. what actually went out.
No. On request we run the architecture entirely as a processor. You get a tenant, a DPA, technical and organizational measures — the same as with any external IT provider, just for AI.
Half day (3.5 hours) or full day (7 hours), on-site in Erfurt, Leipzig, Halle, Jena, Dresden, Bayreuth, or remote. Content is agreed in writing up front; the binding quote follows the intro call.