Skip to content
Sonomos Docs

Sonomos for financial services

A quickstart for relationship managers, financial advisors, analysts, operations, compliance, fraud / AML, treasury, and fintech engineering who want to use AI tools without exposing non-public personal information (NPI), cardholder data, or material non-public information (MNPI).

Who this is for

You’re at a bank, broker-dealer, RIA, insurance carrier, fintech, or payment processor and you or your colleagues are using AI tools for:

  • Drafting client correspondence and personalized communications
  • Summarizing client meetings and call notes
  • Explaining account behavior, transactions, or product features to clients
  • Building internal training material and policies
  • Researching financial products, regulatory questions, market commentary
  • KYC, AML, and fraud-investigation narratives
  • Ops workflows — reconciliations, exception explanations
  • Engineering work on internal tools

Each of these can pull NPI or sensitive financial data into a prompt that leaves your controlled environment.

What’s at stake

Several regulatory regimes simultaneously apply:

  • GLBA Safeguards Rule — requires technical and administrative safeguards for NPI. The 2023 amendments add specific access-control and monitoring obligations.
  • PCI-DSS — any system that processes, stores, or transmits cardholder data is in scope. AI tools that receive a PAN pull themselves into your CDE.
  • SOX — public-company internal controls extend to AI usage by finance, treasury, FP&A, accounting, IR, and legal staff.
  • FINRA / SEC — supervision and record-keeping obligations don’t pause for AI tool use. Firm communications policies must address it.
  • Reg FD — disclosing MNPI to an AI vendor that uses prompts for training is a potential Reg FD problem.

Sonomos addresses the technical-control dimension. The supervision, policy, and disclosure dimensions need their own attention.

The information Sonomos detects

Sonomos’s strongest detector set is its financial set — every detector has a validated checksum:

  • Cardholder data — credit card numbers (Luhn-validated)
  • Bank account data — IBANs (MOD-97), SWIFT/BIC, US bank routing numbers (ABA)
  • Personal identifiers — SSNs (with SSA exclusion rules), EINs, tax IDs, driver’s licenses, passport numbers
  • Investor / customer identifiers — names, DOBs, addresses, emails, phones
  • Health data — relevant for life / disability / LTC insurance underwriting

See the Detector Reference for the complete list.

ChannelWhat to install
Web AI chat (Claude.ai, Gemini, Grok, ChatGPT)Browser extension (Teams plan)
Web-based internal AI toolsBrowser extension
Bloomberg AIM, Refinitiv, or other terminal AIComing: Sonomos Desktop
Native Office CopilotComing: Sonomos Desktop
Fintech engineering / SRE staff using Claude CodeAdd Canary

Setup checklist

  1. Choose Teams from day one. For firms with regulatory exposure, admin policy enforcement and per-member detection statistics aren’t optional. See Plans overview and Teams pricing.

  2. Install the browser extension firm-wide. For larger deployments, see the planned Sonomos Desktop release for MDM-managed deployment via Jamf / Intune / Workspace ONE.

  3. Pin Cloak on for Claude.ai, Gemini, and Grok via Teams admin policy.

  4. Pin Send Guard to “Modal always” for desks that routinely handle NPI or cardholder data: KYC, AML, lending, fraud, ops, and any consumer-facing role.

  5. Pin Send Guard to “Banner + Modal” firm-wide as the floor for everyone else.

  6. Schedule monthly compliance reports to the qualified individual (GLBA), the QSA / compliance officer (PCI-DSS), and the SOX coordinator (where applicable).

  7. Update communications and supervision policies to reference AI tool use and Sonomos as a technical control. Coordinate with the CCO and Reg FD policy owners.

Workflows, step by step

Drafting client communications

  1. Open Claude.ai. Cloak is pinned on (admin policy).
  2. Paste the draft including the client’s name, account number, balance, and recent transactions.
  3. Cloak rewrites NPI to stable placeholders ([CLIENT_NAME], [ACCOUNT_NUMBER], [BALANCE]) before the prompt is transmitted.
  4. The model returns a polished message referencing placeholders.
  5. Substitute the real values back in your CRM or email tool — not by pasting the AI output verbatim.

Summarizing a client meeting

The bread-and-butter wealth management workflow:

  1. Paste meeting notes or transcript into Claude.ai or Gemini.
  2. Client names, account numbers, and SSNs are masked by Cloak.
  3. The model produces a summary you can put into the CRM under the client record.
  4. The masked output is safe; substituting real client identifiers when you save the summary to the CRM keeps the record intact.

Explaining transactions or account behavior

  1. The strongest workflow: ask for the explanation in general without account-specific details. AI can answer “why might an account show a duplicate ACH credit?” without seeing the account.
  2. If you do paste specific transaction lines, Cloak masks the PAN, account numbers, and routing numbers.
  3. Send Guard’s modal will catch any credit card number that slips into a ChatGPT prompt and prevent submission.

KYC and AML investigation narratives

  1. Paste the customer profile elements you need explained.
  2. Names, DOBs, addresses, and identification numbers get masked.
  3. The AI’s analysis references placeholders — copy back into your case management system with real values.
  4. For Modal-mode-enforced desks, every high-severity match (SSN, full account number) gets a deliberate review step.

Fintech engineering on the codebase

For engineers building internal tools or working on production systems:

  1. Add Canary to Claude Code. Two slash commands; no account required.
  2. Canary counts and categorizes PII you’ve shared with Claude across sessions — APIs keys, customer data in stack traces, test card numbers that turn out to be real.
  3. Use the weekly /canary:leaked review to spot patterns that should change.
  4. For browser-based AI (Claude.ai, ChatGPT), the firm browser extension still applies.

Special considerations

Cardholder data and PCI-DSS scope

If your business processes card payments:

  • Set Send Guard to “Modal always” for any role that touches CHD.
  • Acceptable-use policy must prohibit submitting CHD to AI tools, even with Cloak. Cloak is a safety net for accidental disclosure, not a license.
  • Compliance reports document the control’s operation for QSA review.
  • See the PCI-DSS guide for the full scoping and control mapping.

MNPI and Reg FD

Sonomos catches identifiers, not the substance of MNPI. For investment banking, equity research, treasury, FP&A, and corporate development:

  • Written policy must prohibit pasting financial drafts, deal documents, and unannounced earnings details into any AI tool regardless of detection.
  • Awareness training should make clear that Sonomos catches names and identifiers, not concepts.
  • See the SOX guide for the public-company control framing.

Insurance underwriting and PHI

For life, disability, and LTC carriers handling health information:

  • Detect-and-mask treats medical identifiers (MRN, DEA, NPI, diagnoses) as high-severity.
  • Add the HIPAA guide if you’re a covered entity under your business model.

Pitfalls to avoid

  • Bloomberg / Refinitiv / FactSet AI — terminal AI features run inside native apps. Not visible to the browser extension. Treat MNPI inputs with care; await Sonomos Desktop.
  • Native Office Copilot. Out of scope for the extension. Major risk for finance teams that work in Excel and Word.
  • Internal “ChatGPT clones” built on Azure OpenAI, AWS Bedrock, etc. — these run in your tenancy and arguably reduce some risk, but data still leaves the controlled environment. Send Guard activates automatically; consider full Cloak integration via the Teams roadmap.
  • Shared workstations on trading floors. Sign in is per-account; don’t rely on the device’s prior state.
  • Voice transcription of client calls or trading commentary. Otter and built-in OS dictation are not yet covered.

Documenting the control

Every regulator-facing role benefits from being able to point at concrete artifacts:

  • GLBA-qualified individual — monthly compliance reports show the AI-tool channel control operating per §314.4(c) and (h).
  • PCI-DSS QSA — same reports with attention to credit card / cardholder data categories.
  • SOX coordinator / internal audit — per-member detection statistics on Teams plans.
  • FINRA / SEC exam — point at admin policy enforcement plus the supervision policy that references it.

See Compliance reports for generation and routing.

GLBA compliance Safeguards Rule walkthrough for financial institutions.
PCI-DSS compliance Keeping cardholder data out of AI tools.
SOX compliance Public-company ICFR considerations for AI tool use.
Teams setup Pinning admin policy and monitoring per-desk detection statistics.