Trust Accounting Under Florida Bar Rule 5-1.1: Where AI Helps, and Where It Will Get You Disbarred

Rule 5-1.1(g) is the part that disbars Florida lawyers. Not the math. The math is straightforward — it is three-way reconciliation between the IOLTA bank statement, the trust ledger, and the client subsidiary ledgers, and it is exactly the kind of reconciliation an agent runs in seconds.

The reason I keep that distinction front of mind is that vendors currently selling "AI trust accounting" to Florida firms are blurring it, sometimes carelessly and sometimes deliberately. If you are a managing partner in Miami-Dade or a practice administrator in Broward handling IOLTA compliance, the blur is your liability exposure, not theirs.

The Florida Bar's annual lawyer regulation reports have consistently shown trust account violations among the leading factors in disbarment proceedings. The bar is not impressed by the sophistication of your vendor stack when it reviews your client ledger.

Here is where I have come to believe the line actually sits.

Three-way reconciliation at month-end

This is the safe zone for automation, and I will be specific about what that means. The Florida Bar's trust accounting rules, codified in Chapter 5 of the Rules Regulating The Florida Bar (Rule 5-1.1 governing compliance and Rule 5-1.2 governing the specific record-keeping and reconciliation procedures), require monthly reconciliation of three documents: the bank statement for the IOLTA account, the trust journal (running total of all receipts and disbursements), and the individual client subsidiary ledgers. The sum of all client ledger balances must equal the trust journal balance, which must equal the reconciled bank balance. Any deviation is a rule violation regardless of intent.

An agent workflow can pull the bank statement feed (most Florida IOLTA-eligible banks offer OFX or CSV exports; smaller community banks outside that group often do not, which is a real integration constraint), cross-reference it against ledger entries in your practice management software — Clio, MyCase, and CosmoLex all expose API endpoints for this — and flag any discrepancy down to the dollar. The workflow does not interpret the discrepancy. It surfaces it, timestamps it, and routes it to a human. That distinction matters for your compliance posture.

In our deployments, a well-constructed agent running this check completes in under two minutes on a 200-client ledger; the same check run manually by staff typically takes two to four hours depending on complexity. That is not a rounding error in a small firm's economics.

Incoming funds classification and ledger assignment

When a check arrives from a client, something has to decide where it goes: into the IOLTA trust account or into the operating account. For an unearned retainer, it goes into trust. For earned fees already invoiced, it goes into operating. For a settlement advance held for a third party, it goes into trust with a specific client ledger entry. Rule 5-1.1(a) governs the prohibition on commingling and the application of trust funds, and the classifications interact with the retainer agreement terms, the fee arrangement type, and sometimes the client's specific instructions.

An agent workflow can be built to flag incoming deposits, pull the relevant retainer agreement, and return a recommended classification with the specific rule provision it is applying. That is useful. It speeds up the intake step and creates an auditable trail of the classification logic for every deposit. What it cannot do is make the final call. The earned-versus-unearned determination requires a lawyer or a supervised fee earner to review the matter status and confirm. This is not a technical limitation of current models. It is a professional responsibility requirement under Rule 4-5.3 (supervision of nonlawyer assistants), which Florida Bar Ethics Opinion 24-1 extended to generative AI tools in January 2024 by treating AI assistants as nonlawyer assistance subject to the same supervisory duties.

I have seen one vendor demo in Coral Gables — a generic AP-automation tool repositioned for legal — that showed the system auto-posting incoming funds to trust or operating based on its own classification output. No human confirmation step in the workflow. That configuration creates a Rule 5-1.1(a) commingling exposure every time the model misreads a retainer structure, which happens more often than the demo suggests.

Disbursement authorization and the pre-approval gate

Disbursements from trust are where the dollar amounts get large and the risk concentrates. The commingling and misappropriation prohibitions in Rule 5-1.1(a), together with the record-keeping requirements in Rule 5-1.2, mean a lawyer cannot make a disbursement that creates a negative balance in any client's subsidiary ledger — disbursing one client's funds to satisfy another's obligation is the textbook misappropriation case. An agent can enforce this mechanically: before any disbursement is posted, check the client ledger balance, compare it to the requested disbursement amount, and block the transaction if the math does not clear. This is a pre-authorization gate, and it is exactly the kind of deterministic rule-enforcement that should be automated.

What should not be automated is the approval of the disbursement itself. In Florida personal injury and real estate practices specifically (both heavily concentrated in South Florida), trust disbursements at closing or settlement can run into six figures. The workflow should require an attorney signature or a confirmed two-factor authorization before the disbursement instruction is issued to the bank. The agent handles the balance check and the ledger entry. A human handles the authorization. This is a win-win architecture: faster processing, cleaner audit trail, human accountability preserved exactly where the bar requires it.

Rule 5-1.1(g) dispute holds and client fund disputes

This is where I stop recommending any automation at all, and I want to be direct about why. Rule 5-1.1(g) governs what happens when a lawyer and client have a dispute over funds held in trust. The rule requires the lawyer to hold the disputed portion in trust, promptly disburse the undisputed portion, and notify the client. The sequencing of those steps, the characterization of what is disputed versus undisputed, and the timing of the notice all require judgment that is specific to the facts of the dispute and the terms of the fee agreement.

The vendor counterargument here deserves a direct answer. Sophisticated vendors will say their tool only generates a draft letter and a recommended apportionment, with mandatory attorney sign-off, which is materially the same work a paralegal does today under Rule 4-5.3 supervision. The distinction is real and worth stating. A paralegal who has worked in trust accounting for five years has internalized the bar's interpretive guidance, recognizes a fact pattern that triggers a non-obvious rule outcome, and knows when to walk into the partner's office and flag a problem rather than draft around it. A current model produces a confident draft regardless of whether the fact pattern is novel, and the supervising attorney's review is biased by the draft already on the page. That is the asymmetry. Decision-support framing does not eliminate it; it just relocates it to the review step, which is exactly where bar discipline shows attorneys fail.

The Florida Bar Trust Accounting Manual addresses Rule 5-1.1(g) situations explicitly and gives examples where the correct action is counterintuitive if you are just reading the rule text. An agent that flags a ledger anomaly and generates a draft letter is useful as a starting point. An agent that classifies whether a dispute triggers Rule 5-1.1(g) hold requirements, or that decides how to apportion the disputed amount, is creating disbarment risk.

The annual disciplinary data bears this out. Trust account violations that result in disbarment are almost never pure math errors. They are characterization errors, timing errors, and notification failures, all of which require human judgment at every step.

Vendor audit trail and bar examination readiness

The Florida Bar does not audit trust accounts on a fixed schedule, but it does examine records during grievance proceedings and in response to overdraft notifications from banks. Approved financial institutions holding Florida IOLTA accounts are required under Rule 5-1.2 to report trust account overdrafts directly to the bar. When an examination happens, the attorney must produce reconciliation records, client ledger histories, and documentation of the reconciliation process itself.

An agent workflow that runs reconciliation on a consistent schedule and logs every run, every flagged discrepancy, and every human resolution creates a better audit trail than most manual processes produce. Clio's trust accounting module, for example, generates timestamped reconciliation reports. An agent layer on top of that can enforce the monthly reconciliation cadence and generate a signed-off report PDF that gets stored in the matter file. That is genuinely useful for examination readiness, and it is a concrete, conservative use case that every Florida firm running trust accounts should have operational.

The thing nobody is saying clearly enough in vendor presentations is that the audit trail has to show human approval at the classification and disbursement steps. An audit trail that shows "system classified, system disbursed" is not a compliance document. It is evidence of a supervision failure.

Four questions to pressure-test any trust accounting vendor

Before you sign a contract with any platform offering automated trust accounting for a Florida Bar practice, ask these four questions and require written answers in the contract exhibits:

1. At which specific steps in your workflow does the system require human attorney confirmation before any ledger entry or disbursement is finalized, and how is that confirmation logged?

2. How does your system handle a Rule 5-1.1(g) dispute scenario — specifically, does it ever auto-classify or auto-hold funds, or does it stop and route the entire situation to attorney review?

3. What is your error rate on earned-versus-unearned retainer classification across different fee agreement structures, and can you provide that data from production deployments at Florida Bar member firms?

4. If your system produces a misclassification that results in a commingling violation, what indemnification does your contract provide, and has that provision ever been tested in a Florida grievance proceeding?

If the vendor cannot answer questions one and two with specific workflow documentation, the product is not ready for Florida IOLTA compliance use. Questions three and four tell you whether they have built this for Florida lawyers or retrofitted a general accounting tool and added the word "trust" to the marketing page.

The math can be automated safely. The judgment cannot, and every Florida attorney who signs with a vendor that conflates those two things is carrying a disbarment risk the vendor does not share.