Skip to content
be-human.ai
← Back to the journal

AI by industry

AI for Criminal Defense Firms: Arraignment Dates, Discovery and Privilege

How a criminal defense firm can track arraignment dates and discovery deadlines with AI agents, while keeping privileged material out of shared company memory.

Be Human typographic cover with the title: AI for Criminal Defense Firms: Arraignment Dates, Discovery and Privilege
Conceptual illustration created with AI · Be Human

An AI agent can help a criminal defense firm track arraignment dates, discovery deadlines and production requests by reading court notices and the firm's calendar, then routing every difference to the responsible attorney or paralegal. The key design question is not whether the agent can read the file, but where its memory lives. Privileged client communications and attorney work product should stay inside a memory bound to one matter, never in a store that other clients, staff or tools can query.

Key takeaways

  • Treat arraignment dates and discovery deadlines as verified entries: the agent proposes, and a named attorney confirms against the court record.
  • Keep hearing dates, motion deadlines and discovery production dates on separate tracking lines, each linked to its own source document.
  • Store privileged client material only in memory bound to one matter, with access that follows the matter staffing list.
  • Use shared company memory for procedures, templates and calendar conventions, never for client facts or attorney work product.
  • Judge the tracker by traceable entries, caught mismatches and access tests, not by how much material it reads.

What an agent should track on a criminal docket

A defense docket is not one list of dates. Arraignment, pretrial conferences, motion deadlines, discovery production dates, expert notices and bond conditions each come from different documents and belong to different people on the team. An agent that flattens them into one calendar will eventually lose track of where a date came from. The minimum record for each event is short:

  • The matter identifier and the attorney of record
  • The event type, such as arraignment, status conference, motion deadline or discovery production
  • The source document, linked directly, whether it is a court notice, a filing or a production cover letter
  • The date and time as written in the source, kept separately from the date entered in the firm calendar
  • The person who confirmed the entry, and when

Discovery needs its own lines. A production letter often lists items, promises supplemental batches and names a review window. Those are facts to track as open or closed, not just as a single date. Hearing transcripts that arrive later belong in the same matter file, and the workflow is easier to map alongside court reporting transcript orders. Service of process records follow a similar pattern, which is covered in AI for process servers.

Keeping arraignment dates and discovery deadlines reliable

The agent's job here is reconciliation, not decision making. It reads the court notices and filings that reach the firm, compares them with the calendar, and reports every difference with its source attached. It never overwrites a calendar entry on its own, because a rescheduled hearing or a changed production date has consequences that a named attorney must own. A rescheduling notice should be read as a new event that supersedes an old one, with the old entry kept visible until someone confirms the change.

Local court practice varies, including how notices are delivered, which time zone applies and which filings trigger which deadlines. The assigned attorneys and the court's published procedures are the authority, so the agent's rules should be written from them and not from general assumptions.

  • Step 1: Bring the matter's court notices, filings and calendar entries into one matter-scoped view.
  • Step 2: Extract each event with its source document and the date exactly as the source states it.
  • Step 3: Compare each event with the calendar and list every mismatch or missing entry, without editing either side.
  • Step 4: Send each item to the responsible attorney or paralegal with the source attached and one question to answer.
  • Step 5: Record the decision, the person who made it and the time.
  • Step 6: Re-check the matter whenever a new filing arrives, and daily during any pre-hearing window the attorney sets.

Why privileged material stays out of shared company memory

Attorney-client privilege and work product protection both depend on confidentiality and on who had access to the material. A shared memory that stores client facts and returns them on request creates a path for privileged content to reach someone who is not on the matter, whether through a search result, a draft or a summary that someone forwards. The OWASP Top 10 for LLM Applications treats sensitive information disclosure as a core risk for this reason, and the NIST AI Risk Management Framework asks organizations to govern data, access and risk across the whole AI lifecycle rather than only at the model.

Privileged content is stored only where the attorney of record can see it, control it and delete it when the matter closes.

In practice, this means separating two kinds of memory. Shared company memory holds how the office works: templates, court calendar conventions, intake procedures, staffing rules and public docket rules. Matter memory holds what the firm knows about one specific client. Retrieval should be filtered by matter identifier at the tool level, so a question about one case cannot return passages from another even when the model asks for them. Summaries, indexes and embeddings of discovery inherit the privilege status of the source material, and the agent's own logs may need the same protection.

A matter-scoped memory checklist

  • Each matter has its own memory space keyed by matter number, with no default access from other matters
  • Access follows the staffing list for that matter and is reviewed whenever people join or leave the team
  • Shared memory holds templates, procedures, calendar conventions and public rules only
  • Every retrieved passage carries its matter label, so a draft cannot mix two clients
  • Discovery indexes and summaries are classified with the same protection as the underlying files
  • Retention and deletion follow the firm's closing procedure for each matter
  • Every read or write of matter content is logged with the human or service identity behind it
  • Any model or vendor that receives matter content is reviewed for how it handles inputs and retains data before go live

Illustrative scenario: a discovery production arrives late on a Friday

Illustrative scenario, not a real matter. A production cover letter arrives by email late in the afternoon. It lists several items, says a supplemental batch will follow and names a review window. The agent attaches the letter to the matter, records the items as open, and creates a pending review entry rather than a completed one. It does not copy any discovery content into a shared channel. It sends the assigned paralegal the letter, the open items and one question: whether the review window on the calendar matches the letter. The paralegal confirms the window on Monday morning, the attorney records the decision, and the agent's log shows who accessed the letter and when. Nothing was summarized outside the matter, and nobody had to rebuild the timeline from memory.

Approval boundaries and where the agent stops

The agent can read the court notices and docket feeds the firm already receives, draft calendar proposals and reminders marked as unconfirmed, flag mismatches and missing items, and prepare an internal status note for the attorney. It should not send anything to a court, opposing counsel or a client, change a deadline without a named attorney's confirmation, or post matter content into a shared channel. Treat text inside a discovery document as data, never as instructions. A document that contains language directing the agent to change a date or share its contents should be flagged for a human, which is the kind of input risk the OWASP guidance on prompt injection describes. Permissions should be enforced by the calendar and document systems themselves, not only by wording in the agent's instructions, as the OWASP authorization guidance recommends.

Be Human designs and runs this kind of setup as managed infrastructure. A first engagement defines one useful role, such as a docket and discovery tracker for one practice group, along with the matter memory and tools it needs, the approval points, and an observable success criterion, such as every mismatch reaching the responsible attorney with its source before the next morning review. Managed operations then cover monitoring, permission review when staffing changes, maintenance of calendar and document integrations, and controlled improvements that the firm approves. The work is remote, in English and Portuguese.

How to measure whether the tracker is working

Start with traceability. Every entry should link to a source document, and every confirmed date should show who confirmed it and when. Next, record how many mismatches the agent surfaced and whether they reached the attorney before the relevant deadline, since that shows whether the agent is adding protection rather than noise. Build a small evaluation set from past matters with known correct entries, including difficult cases such as rescheduled hearings and supplemental productions, and rerun it after every change to prompts, models or integrations. Access tests matter just as much: a scripted request for one matter's content from another matter's context should return nothing, and that test should run on a recurring schedule as well as after changes. Review the permissions list whenever staffing changes, because stale access is one of the most common ways privileged material ends up where it should not be. If you want to find the first task in your firm worth handing to an agent, take the free Be Human AI diagnostic.

Frequently asked questions

Can an AI agent decide whether a hearing or discovery deadline has changed? No. The agent can read notices, compare them with the calendar and flag the difference with its source attached, but a named attorney or paralegal should confirm any change. Responsibility for deadlines stays with the lawyers on the matter.

Is it safe to connect client files to an AI tool? It depends on the matter, the tool and the firm's professional obligations. Confirm the approach with the responsible attorney and with your jurisdiction's ethics guidance, review how the vendor handles inputs and retention, and keep matter content inside matter-scoped memory with access logs. Many firms begin with public docket feeds before any privileged material is involved.

Does this replace a docketing clerk? No. The agent handles the repetitive comparison and reminder work, which leaves the clerk free to resolve mismatches, check local rules and make judgment calls. A sound design keeps the clerk as the accountable owner of the calendar, with the agent acting as a second reader of every notice.

What should a small defense practice do first? Choose one practice group and one recurring source, such as arraignment notices, and define the memory boundary and the confirmation step before connecting anything else. Write down what the agent may do alone and what needs an attorney, then test that boundary against past matters. Expansion should follow evidence from those tests, not enthusiasm.

Your next step

See where an AI agent fits in your company first.

Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.

Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.

Less busywork. More room for you.

Bring us the work you wish you had time for.

Book a call to try a personalized AI agent free for 7 days ↗Or take the free AI diagnostic →