In May 2026 a FINRA AWC settled with a mid-Atlantic broker-dealer over a familiar pattern: a registered representative had recommended an IRA rollover for a 62-year-old separating from a Fortune 500 employer; the firm's Reg BI file contained a recommendation memo, an account-opening package, and signed disclosure delivery acknowledgments — but lacked any documented consideration of the four reasonably available alternatives (leave in plan, roll to new employer plan, roll to IRA, take cash). The AWC named the registered representative, named the firm, imposed a fine, and required additional supervision and remediation. The kicker: the recommendation memo had been drafted in under three minutes by an AI tool from a Zocks meeting transcript. The AI had not failed; the supervisory architecture around the AI had failed. This lesson re-reads Reg BI's four obligations — Disclosure, Care, Conflict, Compliance — through an AI lens and builds the defensible 2026 Reg BI file every advisor, every CCO, and every OSJ supervisor needs when AI touches any part of a recommendation.

Reg BI Codification and the Four Obligations

Regulation Best Interest is codified at 17 CFR §240.15l-1, adopted by the Securities and Exchange Commission in June 2019 and effective June 30, 2020. The rule applies to broker-dealers and their associated persons when making a "recommendation" of any securities transaction or investment strategy involving securities to a retail customer. "Retail customer" is defined broadly to include natural persons receiving recommendations of securities for their own personal, family, or household purposes. The four obligations — sometimes called the four "component obligations" — that comprise the Reg BI best-interest standard are codified at §240.15l-1(a)(2)(i) through (iv), and the framework is the spine of every AI-and-recommendation analysis in 2026.

Disclosure Obligation — §240.15l-1(a)(2)(i)

The broker-dealer must, prior to or at the time of the recommendation, provide the retail customer in writing full and fair disclosure of (a) all material facts relating to the scope and terms of the relationship, including the capacity in which the broker-dealer is acting, and (b) all material facts relating to conflicts of interest associated with the recommendation. Form CRS (Customer Relationship Summary), the customer-relationship-summary required of broker-dealers and dually-registered advisers, carries much of the relationship-scope disclosure. The recommendation-specific conflicts disclosure rides on top.

Care Obligation — §240.15l-1(a)(2)(ii)

The broker-dealer must exercise reasonable diligence, care, and skill to (a) understand the potential risks, rewards, and costs of the recommendation, (b) have a reasonable basis to believe the recommendation could be in the best interest of at least some retail customers, (c) have a reasonable basis to believe the recommendation is in the best interest of the particular retail customer based on that customer's investment profile, and (d) when recommending a series of transactions, have a reasonable basis to believe the series is not excessive. The Care Obligation is the obligation that absorbs the most AI-related scrutiny in 2026 because it requires documented professional judgment that an AI tool cannot supply.

Conflict of Interest Obligation — §240.15l-1(a)(2)(iii)

The broker-dealer must establish, maintain, and enforce written policies and procedures reasonably designed to identify and at a minimum disclose, or eliminate, all conflicts of interest associated with recommendations to retail customers. Certain categories of conflicts — those creating incentives to place the broker-dealer's interest ahead of the customer's — must be mitigated, and certain practices (e.g., sales contests, sales quotas, bonuses, and non-cash compensation tied to the sale of specific securities or types of securities within a limited period of time) must be eliminated.

Compliance Obligation — §240.15l-1(a)(2)(iv)

The broker-dealer must establish, maintain, and enforce written policies and procedures reasonably designed to achieve compliance with Reg BI. The Compliance Obligation is what couples Reg BI to FINRA Rule 3110 supervision and Rule 4511 recordkeeping — the procedures, the supervisory review, and the retained evidence are how the firm demonstrates Reg BI conformance under examination.

Re-Reading the Four Obligations Through an AI Lens

Each of the four obligations changes operational shape when AI is in the workflow. The substance of the obligation does not change; the evidence the firm needs to produce changes.

AI and Disclosure

When an AI tool drafts the Reg BI disclosure or the Form CRS delivery package, the disclosure obligation runs not on what the AI produced but on what the registered person delivered to the customer. Two failure modes emerge in practice. First, an AI-drafted disclosure that omits a conflict of interest the firm in fact has — for example, a custodian-specific compensation arrangement or a proprietary-product preference — fails the obligation regardless of whether the AI was instructed to include it. Second, an AI tool that updates relationship-scope language after a workflow change (a new custodian, a new fee schedule, a new advisory service tier) without flowing into Form CRS and the recommendation file creates an inconsistency the SEC and FINRA can examine on. The defensive pattern: AI may draft, the human verifies against the firm's actual conflicts and the current Form CRS, the human delivers, and the delivery is logged.

AI and Care

The Care Obligation is the one AI tools touch most directly and the one that absorbs the most enforcement scrutiny in 2026. The four sub-elements — understand risks/rewards/costs, reasonable basis some retail customer, reasonable basis this customer, no excessive series — each require documented evidence. AI summarizes product literature, extracts fee data, and proposes alternatives at order-of-magnitude lower cost; the Care Obligation still requires the registered person's documented consideration. The 2025-2026 FINRA AWC pattern on inadequate rollover documentation turns on this exact point: the AI memo was drafted; the registered person signed; the file did not show the registered person's documented consideration of the reasonably available alternatives. The fix is the alternatives-considered section of the memo, populated by the registered person (not the AI) with the specific reasoning for each alternative.

AI and Conflict of Interest

The Conflict Obligation has two distinct AI implications. First, AI tools themselves create conflicts the firm must identify and disclose — vendor compensation arrangements (revenue-share, referral fees), data-sharing arrangements with third parties, and AI-tool-driven product preferences (e.g., an AI tool the firm uses that surfaces in-house mutual funds more than competitor funds because of training-data composition). Second, AI tools that surface recommendations can amplify or mask existing conflicts — if the firm has an undocumented incentive to prefer one custodian over another and the AI surfaces that custodian's products more often, the firm has a Conflict Obligation failure even if no human consciously biased the output. The WSPs must inventory AI-tool conflicts and the supervisory review must check AI outputs for conflict drift.

AI and Compliance

The Compliance Obligation couples directly to FINRA Rule 3110 (supervision) and Rule 4511 (recordkeeping). The firm must have WSPs reasonably designed to achieve Reg BI compliance, the supervisory architecture must surface AI-touched recommendations for principal review where appropriate, and the recordkeeping must retain the inputs (transcripts, prompts, source documents), the AI outputs, the registered person's edits and consideration of alternatives, and the supervisory signoff. The L1 Ch4 L3 lesson develops Rules 2210/3110/4511 in detail; the L4 Ch3 lessons develop the WSPs at scale. This lesson establishes the foundational frame.

The Defensible 2026 Reg BI File — What Goes In

The single most consequential operational artifact for an advisor or BD in 2026 is the per-recommendation Reg BI file. The file is what an examiner reads to evaluate whether the firm met the four obligations. The file is what the AWC respondent could not produce. The defensible file has six components, each populated by a different kind of input.

Component 1: Recommendation Memo

A written record of the recommendation: what is being recommended, to whom, on what date, by which registered person. AI tools can draft this from a meeting transcript (Zocks, Jump, FinMate, Sybill, Zeplyn) in seconds. The memo must accurately describe the recommended action — the specific securities, the specific account, the specific transaction or strategy.

Component 2: Alternatives Considered

The single most often missing component in the AWC pattern. The Care Obligation requires the registered person to have considered reasonably available alternatives to the recommendation; the file must show what alternatives were considered and why they were not selected. For a rollover recommendation, the four canonical alternatives are: leave in plan, roll to new employer plan, roll to IRA, take cash. For a fund recommendation, the alternatives may be peer funds, share-class variants, and ETF alternatives. For an annuity recommendation, the alternatives include lower-cost product variants, no-annuity (taxable / IRA), and laddered structures. The AI tool can list alternatives; the registered person must document the consideration with specific reasoning.

Component 3: Costs Comparison

An itemized comparison of the costs of the recommendation against the costs of the alternatives. For rollovers, the share-class comparison between the plan and the IRA destination is the canonical example. For mutual fund or ETF recommendations, the expense ratios, transaction costs, and any wrap-fee or advisory-fee differentials. For annuities, the surrender charges, mortality-and-expense fees, rider costs, and surrender-period exposure. AI tools collapse the cost-extraction time; the registered person verifies and the comparison is in the file.

Component 4: Client-Specific Rationale

The reasonable-basis-for-this-customer element under §240.15l-1(a)(2)(ii)(C). The file must show why the recommendation is in the best interest of this customer based on this customer's investment profile (age, financial situation, investment experience, investment objectives, risk tolerance, liquidity needs, time horizon, tax considerations, other holdings). AI can pull the profile data from the CRM (Wealthbox, Redtail, Salesforce FSC); the registered person writes the rationale and ties it to the profile elements.

Component 5: Reviewer / Principal Signoff

The supervisory layer. FINRA Rule 3110 requires reasonable supervision; the firm's WSPs designate which recommendation categories require principal review pre-execution. For BDs, certain transactions require principal approval. The signoff is recorded — timestamped, attributed, retained — and joins the file. The L4 Ch3 lesson on principal review of AI-drafted communications under Rule 2210 builds the at-scale workflow.

Component 6: Retained Inputs and AI Artifacts

The transcripts, prompts, AI tool / model version, AI outputs, human edits, and the production chain. Retained under FINRA Rule 4511 (three years for BDs) and SEC Rule 204-2 (five years for RIAs, first two easily accessible). The records are what allow reconstruction of how the recommendation was produced if an examiner asks.

The May 2026 FINRA AWC Pattern of Inadequate Rollover Documentation

Through 2025 and into 2026, FINRA continued to bring Letters of Acceptance, Waiver, and Consent ("AWCs") against broker-dealers and individual registered representatives for inadequate Reg BI documentation, with rollovers as the dominant fact pattern. The recurring elements across the AWCs: (a) the registered representative recommended a rollover from an ERISA-covered plan to an IRA; (b) the recommendation file contained the recommendation memo but did not show documented consideration of leaving the assets in the plan, rolling to a new employer's plan, or taking a cash distribution; (c) the firm's supervisory review did not flag the missing alternatives-considered documentation; (d) the firm and the registered representative were named, fined, and required to implement additional supervision and remediation.

The 2026 twist — material to this lesson — is that AI tools materially compress the time cost of producing the recommendation memo (Zocks transcript to Jump-drafted memo to Wealthbox activity log in under three minutes) while doing nothing to compress the registered person's documented-consideration time. If the registered person treats the AI memo as the entire file and signs without populating alternatives-considered with specific reasoning, the firm has produced the same inadequate file faster — and the examination exposure is the same. The fix is the registered person's discipline: every AI-drafted rollover memo gets the alternatives-considered section populated by the registered person before signoff, with specific reasoning tied to the client's investment profile.

Rollover Alternatives — Populating the Section

For each of the four canonical rollover alternatives, the documented consideration includes: (1) Leave in plan — what plan investment options exist, what expense ratios apply, what loan provisions exist, what employer-stock NUA election under IRC §402(e)(4) exists, what required-distribution timing applies; (2) Roll to new employer plan — does the new employer plan exist, does it accept rollovers-in, what investment options does it offer, what expense ratios apply, what mega-backdoor Roth eligibility exists; (3) Roll to IRA — share-class comparison (plan vs. IRA destination), advisory fee impact, investment options expansion, distribution flexibility, beneficiary planning, pro-rata rule consideration under IRC §408(d)(2) read with §72(e)(8) and Form 8606 if a backdoor Roth is intended; (4) Take cash — tax impact (ordinary income plus IRC §72(t) 10% additional tax if under 59½ with no SEPP under §72(t)(2)(A)(iv)), liquidity benefit, lost retirement-savings opportunity. Specific reasoning is what survives examination.

AI-Recommendation Types Sorted by Reg BI Exposure

Not every AI-touched workflow produces a "recommendation" in the Reg BI sense. The technology-neutral test: did a recommendation to a retail customer occur, and was AI in the workflow that produced it? Three categories help sort.

High Exposure: AI Drafts the Recommendation Memo

Rollover memos, mutual fund / ETF / annuity recommendations, IRA-to-IRA transfers between custodians (which may or may not be recommendations depending on facts), 529-plan recommendations, business owner retirement plan recommendations. AI draft + registered person signoff = recommendation. Full Reg BI file required. This is the lesson's core operational frame.

Medium Exposure: AI Summarizes Information Used in the Recommendation

Holistiplan 1040 extraction, FP Alpha estate-document extraction, Wealth.com beneficiary extraction, RightCapital plan summary. AI summary feeds the registered person's recommendation; the recommendation is the registered person's. Reg BI file required for the recommendation; AI summary is part of the inputs retained.

Lower Exposure: AI Supports Non-Recommendation Workflows

Meeting prep briefs, follow-up email drafts that don't contain recommendations, CRM activity log entries, IPS draft language. The Reg BI four obligations don't attach in the same way because no recommendation occurred; the Marketing Rule, Rule 2210, Reg S-P, and Rule 4511 may still apply.

The CCO Perspective — Rule 3110 Coupling and Supervisory Architecture

From the CCO or OSJ supervisor's seat, Reg BI's Compliance Obligation couples directly to FINRA Rule 3110 reasonable supervision. The supervisory architecture must: (a) define which recommendation types require principal review pre-execution; (b) define the principal review checklist that includes the alternatives-considered adequacy check; (c) define the AI-tool acceptable-use boundaries (which tools may touch recommendation drafts, which data categories may be input, what verification protocol applies); (d) define the recordkeeping that captures the AI production chain under Rule 4511 and Rule 204-2; (e) define the exception escalation when the AI surfaces a recommendation that doesn't fit the customer profile or appears to amplify a conflict. The L1 Ch4 L3 lesson on Rules 2210/3110/4511 and the L4 Ch3 lessons on WSPs operationalize this; the present lesson establishes the substantive coupling.

Key Takeaways

• Reg BI is codified at 17 CFR §240.15l-1, effective June 30, 2020, applies to broker-dealers and associated persons recommending securities transactions or strategies to retail customers, and comprises four obligations: Disclosure §240.15l-1(a)(2)(i), Care §240.15l-1(a)(2)(ii), Conflict §240.15l-1(a)(2)(iii), Compliance §240.15l-1(a)(2)(iv).
• The 2025-2026 FINRA AWC pattern turns on inadequate rollover Reg BI files — recommendation memo present, alternatives-considered absent. AI tools collapsing the memo-drafting time amplify the exposure if registered persons sign without populating the alternatives-considered section with specific reasoning.
• The defensible 2026 Reg BI file has six components: recommendation memo, alternatives considered, costs comparison, client-specific rationale, reviewer/principal signoff, and retained inputs/AI artifacts (transcripts, prompts, model version, outputs, edits, signoff).
• The four canonical rollover alternatives — leave in plan, roll to new employer plan, roll to IRA, take cash — each get a specific-reasoning entry populated by the registered person, not the AI. Pro-rata rule consideration under IRC §408(d)(2) + §72(e)(8) + Form 8606 belongs in the IRA-rollover alternative when a backdoor Roth is intended.
• AI tools themselves can create Reg BI conflicts (vendor compensation, data sharing, training-data product bias) that the WSPs must identify and the supervisory review must check for conflict drift in outputs.
• Three AI-recommendation exposure tiers: high (AI drafts the recommendation memo), medium (AI summarizes inputs the registered person uses), lower (AI supports non-recommendation workflows). Each tier has a different file requirement.
• The Compliance Obligation couples Reg BI to FINRA Rule 3110 supervision and Rule 4511 recordkeeping, developed at workflow scale in L1 Ch4 L3 and at WSP scale in L4 Ch3.