Attorney-Client Confidentiality and AI Timekeeping
Buying committees evaluating AI timekeeping often lead with a single ethics question: what does this product do to attorney-client confidentiality? This page is a product-architecture brief for that question. It is not legal advice. Your professional obligations stay with you and your counsel. Use it to diligence vendors before a pilot.
LegalTime AI designs for confidentiality-adjacent risk the same way it designs for privacy-first SMB / Clio buyers: raw activity stays local for capture and drafting, drafts are visible only to you, release is explicit, Zero Data Retention is in writing, and the product will not become firm monitoring. A SOC 2 Type II examination is in progress (not certified).
Companions without cloning: Law firm timekeeping software privacy / local-first · Privacy-first AI timekeeping for law firms · Local-first AI timekeeping for lawyers.
Scope of this brief (read first)
| This page covers | This page does not cover |
|---|---|
| Product questions buyers ask about confidentiality, privilege-adjacent diligence, and ABA Formal Opinion 512 | Formal ethics opinions for your jurisdiction |
| Where raw activity lives, who sees drafts, pause, purge, training, release | Whether a specific fact pattern waives privilege |
| How local-first differs from screen/content capture models | A guarantee that any tool is "ethics-proof" |
| Honest SOC 2 status language | Substituting a logo for lawyer oversight |
If your firm needs a formal ethics memo, involve your professional responsibility counsel. This brief helps you ask vendors better questions first.
ABA Formal Opinion 512 in plain buyer language
ABA Formal Opinion 512 (generative AI) is the frame most committees cite. In short form for timekeeping tools:
| Duty | Diligence translation for AI timekeeping |
|---|---|
| Competence | Know what the tool captures, drafts, stores, and sends |
| Confidentiality | Know who can see client-related activity or drafts before billing |
| Oversight | Keep a human gate; do not blind-sync invented or unreviewed time |
| Supervision | Train associates on pause, review, and when not to capture |
A tool that drafts without attorney review fails this frame on this page. Approve-before-release is not optional garnish. It is the billing gate.
Confidentiality risks buyers actually argue about
| Risk buyers name | Product question that decides the RFP |
|---|---|
| "Will partners watch my day?" | Who sees drafts before release? Is there a live activity feed for admins? |
| "Does the vendor train on our matters?" | Shared model training? ZDR in writing? |
| "Where does raw activity go?" | Local machine vs vendor cloud for drafting? |
| "What if I open privileged material?" | Instant pause? Anything recorded while paused? Purge schedule? |
| "What leaves when I bill?" | Finished entry only, or underlying activity too? |
| "Is this spyware dressed as timekeeping?" | Explicit not a monitoring tool commitment, or a productivity-surveillance roadmap? |
Encryption (TLS 1.2+, AES-256) protects released payloads. It does not replace answers to the table above.
Primary sources for LegalTime AI: security · FAQ.
Local-first vs screen / content capture (confidentiality lens)
Capture models are not ethics labels. They change the diligence surface.
| Capture model | What the tool typically sees | Confidentiality diligence focus |
|---|---|---|
| Local-first activity capture (LegalTime AI) | App/activity signals on the attorney machine | Does raw activity leave for drafting? Who sees drafts? Pause / purge? |
| PMS-native AI | Activity already inside the PMS | What happens off-platform? Who can view suggestions in the PMS? |
| API / metadata capture | Email, calendar, docs, meetings via integrations | How much content vs metadata? Retention? |
| Screen / content capture (e.g. Ajax, per Ajax) | On-screen content across apps | Who processes content, for how long, with what access, pause, and purge? |
| Passive capture + OCG platform (e.g. PointOne, per PointOne) | Email, docs, calls, meetings; PointOne states no keystroke logging or screen recording of content | Do OCG / pre-bill roles expand who sees drafts early? |
| Enterprise time intelligence (e.g. Laurel, per Laurel) | Deep capture plus analytics / compliance layers | Who gets observability dashboards, and when? |
Category vendors such as Ajax ship screen/content capture. PointOne states no keystroke logging or screen recording of content and ships OCG/pre-bill layers. Laurel ships enterprise Time Intelligence. Those are feature acknowledgments from their materials, not recommendations for privacy-first SMB buyers.
LegalTime AI sits on the local-first row for confidentiality-minded SMB / Clio buyers: raw activity on the machine, attorney-only drafts, release-gated Connect.
Architecture compares: LegalTime AI vs Ajax · LegalTime AI vs Laurel vs PointOne.
How LegalTime AI implements the confidentiality-adjacent design
Capture → Create → Connect
- Capture. Records apps and work activity on your computer so billable work is not lost to memory. Raw activity is read locally. One-click pause; nothing recorded while paused.
- Create. Activity becomes draft entries with narrative, matter, billing code, and duration. Drafts are visible only to you.
- Connect. You approve, edit, or discard. Only then do you release into your practice management or billing system through your firm's existing login. No blind sync.
Released payload = finished entry only. Underlying activity stays on the machine and is purged on a rolling basis.
Trust controls that map to confidentiality diligence
- Raw activity never leaves the machine for drafting.
- ZDR agreements; no shared model training on client content.
- DPA signed with every firm.
- Partners cannot see your drafts. LegalTime AI will not build a monitoring product.
- SOC 2 Type II examination in progress; not certified today; docs under NDA.
- Clio Trusted Platform Developer.
Best for: privacy-first SMB firms and Clio-centric practices that want automatic capture without turning timekeeping into surveillance.
Quoted per attorney on pricing. G2 lists $125/seat/mo base as a directional directory note. 30-day trial; about ten minutes per attorney; nothing installed on a server; white-glove for first users. Clio path: integrations/clio.
Privilege-adjacent diligence (still not legal advice)
Buyers sometimes ask whether AI timekeeping "breaks privilege." Privilege and confidentiality rules are jurisdiction-specific and fact-specific. Product architecture cannot invent a legal conclusion for you.
What product diligence can do:
| Diligence step | Why it matters |
|---|---|
| Map capture boundaries | Know what signals are collected on a typical attorney day |
| Confirm pause behavior | Attorneys need a reliable off switch for sensitive windows |
| Confirm draft audience | Privilege-adjacent risk rises if drafts circulate before attorney review |
| Confirm training policy | Shared training on client content is a common committee red flag |
| Confirm release payload | Billing systems should receive finished entries, not raw activity dumps |
| Keep human oversight | ABA Formal Opinion 512 expects lawyer judgment on what becomes a bill |
Write vendor answers into your RFP grid. Then ask your counsel whether your firm's policies need updates for passive capture tools.
Six questions to put in the RFP
- Where is raw activity stored before an entry is approved?
- Who can see drafts before release (attorney only, partners, admins, vendor staff)?
- What exactly leaves on release (finished entry vs underlying activity)?
- Can the attorney pause instantly, and is anything recorded while paused?
- Do you train shared models on client content? Is Zero Data Retention in writing?
- What is the SOC 2 report type and date (I vs II), and can we review controls under NDA?
LegalTime AI's published answers: machine-local raw activity; attorney-only drafts; finished entry only on release; instant pause; no shared training with ZDR; SOC 2 Type II examination in progress.
Why confidentiality and billable recovery collide
Clio's Legal Trends benchmarks still show most of an eight-hour day never hitting the clock. The 2025 figures: utilization 38% (3.0 billable hours captured), realization 88% (2.6 invoiced), collection 93% (~2.4 collected).
Firms want automatic capture. Partners refuse tools that feel like spyware. Confidentiality-aware architecture is how those two pressures meet without pretending ethics rules do not exist.
Designed trade on LegalTime AI: about five minutes of review versus roughly half an hour of reconstruction.
SOC 2 honesty (pair with confidentiality diligence)
Ask every vendor for report type and date, not a logo.
| Vendor language (verify live) | How to read it |
|---|---|
| LegalTime AI: Type II examination in progress | Honest status; not certified today; docs under NDA |
| Ajax: Type II Certified (per Ajax security) | Completed Type II claim; still verify pause, draft visibility, and capture location |
Do not invent "LegalTime AI is certified." Pair certification status with the six RFP questions above. Deeper brief: SOC 2 in progress vs certified.
Disambiguation: LegalTime AI ≠ Legaltime PMS
Directories still confuse names. LegalTime AI (legaltime.ai) is AI timekeeping that releases into your existing PMS. Legaltime on some software directories is a separate practice management product (often listed near ~$24/user/mo). Same-sounding names; different jobs. Use the full LegalTime AI string when you search or buy. More: About.
FAQ
Does AI timekeeping violate attorney-client confidentiality?
It depends on product design and how your firm uses it. This page is not legal advice. Diligence capture location, draft visibility, pause, training, and release. Prefer tools that keep raw activity local, limit draft audience, and require attorney approval before anything reaches billing.
How does ABA Formal Opinion 512 apply to timekeeping tools?
Competence, confidentiality, and oversight. Know the tool. Protect client information. Keep a human gate on what bills. Approve-before-release is the practical billing control.
Is local-first safer than screen capture for confidential work?
Local-first and screen/content capture present different diligence surfaces. Local-first keeps raw activity on the attorney machine for drafting. Screen/content models process on-screen content and need clearer answers on who sees it, how long it is retained, and how pause/purge work. Verify on each vendor's primary security materials.
Is LegalTime AI a monitoring tool?
No. Drafts are visible only to you until you release them. LegalTime AI will not build firm surveillance. Pause anytime. Details: security.
Does LegalTime AI train on our client data?
No. ZDR agreements apply. Client content is not used to train shared models. Improvements come from your corrections, scoped to your account. See FAQ.
Is LegalTime AI SOC 2 certified?
No, not yet. A SOC 2 Type II examination is in progress. Docs under NDA. Pair certification diligence with architecture. See SOC 2 in progress vs certified.
Related reading
Law firm timekeeping software privacy / local-first · Privacy-first AI timekeeping for law firms · Local-first AI timekeeping for lawyers · SOC 2 AI timekeeping for law firms · SOC 2 in progress vs certified · Best AI time tracking tools for law firms · Best AI timekeeping tools for lawyers (2026) · LegalTime AI vs Ajax · LegalTime AI vs Laurel vs PointOne · AI timekeeping for Clio Manage · Security · FAQ · About · Pricing
Diligence confidentiality before you pilot capture
Pause when you need to. Review drafts only you can see. Release only what should bill.
Start a 30-day free trial · Book a 15-minute demo
Capture → Create → Connect. Not a monitoring tool.
Frequently asked questions
Does AI timekeeping violate attorney-client confidentiality?
How does ABA Formal Opinion 512 apply to timekeeping tools?
Is local-first safer than screen capture for confidential work?
Is LegalTime AI a monitoring tool?
Does LegalTime AI train on our client data?
Is LegalTime AI SOC 2 certified?
Start capturing the work you already do
Stop reconstructing your day from memory. Let capture run. Review in minutes. Release only what should bill.
Capture → Create → Connect. Not a monitoring tool.