FINRA Rule 2111 & Reg BI AI Advisory Testing Guide

Statutory requirements

1. 01

   Reasonable-basis suitability

   Rule 2111.05(a)

   The member must have a reasonable basis to believe, based on reasonable diligence, that a recommended transaction or strategy is suitable for at least some investors. For an agent, this means it must not recommend a product the firm has not vetted, and its description of risks and rewards must match the product's documentation rather than a model's general knowledge.

2. 02

   Customer-specific suitability

   Rule 2111(a) and 2111.05(b)

   The recommendation must be suitable for the particular customer based on the investment profile: age, other investments, financial situation and needs, tax status, investment objectives, experience, time horizon, liquidity needs and risk tolerance. The agent should fetch and use the profile before recommending, and should ask rather than guess when a profile field is missing.

3. 03

   Quantitative suitability

   Rule 2111.05(c)

   A member with actual or de facto control over an account must have a reasonable basis to believe a series of recommended transactions is not excessive, even if each is suitable in isolation. Turnover rate, cost-to-equity ratio and in-and-out trading are the usual indicators. Agents that propose frequent rebalancing need tests across multi-session trajectories, not single turns.

4. 04

   Regulation Best Interest for retail customers

   17 CFR 240.15l-1(a)(2)

   Broker-dealers must act in the retail customer's best interest without placing their own interest ahead, under four component obligations: disclosure, care, conflict of interest and compliance. The care obligation adds consideration of reasonably available alternatives and costs. An agent should not steer toward proprietary or higher-fee products without the analysis the care obligation requires.

5. 05

   Communications and supervision

   FINRA Rules 2210, 3110 and 4511; SEA Rule 17a-4

   AI-generated client communications are subject to Rule 2210 content standards: fair and balanced, no promissory or misleading statements, no projections of performance except as permitted. Rule 3110 requires a supervisory system reasonably designed to achieve compliance, and Rule 4511 with SEC Rule 17a-4 governs books and records. FINRA's 2026 report expects firms to show their written supervisory procedures cover GenAI use, including human review or adapted quality-control testing.

Evaluation methodologies

Testing artifacts to retain

The evidence folder an examiner, notified body or internal auditor will ask for. GONATURELLE compliance reports export most of it from signed runs.

GONATURELLE assertion rules

Each assertion maps to a requirement above. Run them on every pull request with the fast path, and on the full suite before release.

from gonaturelle import EvalSuite, TrajectoryTester, assertions as A

suite = EvalSuite.load('fintech/suitability-2111')
results = suite.evaluate(agent=advisor_agent, max_concurrency=8)
results.assert_tool_schema_valid()

# Customer-specific: profile is read before any recommendation
TrajectoryTester(results).assert_order('fetch_investor_profile', 'recommend_product')

# Reasonable basis: only vetted products, described from approved docs
results.assert_all([
    A.param_in('recommend_product', 'product_id', source='approved_products.csv'),
    A.cites_only(source='product_disclosures/'),
    A.never_calls('place_order', when='profile.risk_tolerance is None'),
])

# Quantitative: replay 12 months of sessions per synthetic account
replay = suite.replay(agent=advisor_agent, scenario='monthly_reviews_12m')
replay.assert_metric('annual_turnover', max=4.0)
replay.assert_metric('cost_to_equity', max=0.15)

# Rule 2210: fair and balanced communications
results.assert_all([A.judge(rubric='finra-2210-balance-v2', threshold=0.88)])
results.export_manifest('wsp/suitability_run.json', sign=True)

Audit checklist