What banks can market and what the rules say
Financial advertising rules across all five of leapbuzz's priority markets share a structural logic: the more complex, risky, or opaque the product, the more prescriptive the advertising constraints. The practical consequence for marketing teams is that the channel, format, and copy approval workflow differs materially by product type and jurisdiction.
| Market | Controlling instrument | Key 2025-2026 change | Equal-prominence rule |
|---|---|---|---|
| Singapore | MAS Guideline FSG-03 (effective 25 March 2026); MAS FAA-N16 for ads naming specific investment products | FSG-03 brought influencers, affiliates, and agencies into compliance perimeter. Board accountable for third-party campaigns. May 2025 consultation proposed removing exclusions for accredited/institutional investor ads. | Yes. FSG-03 requires addressing format limitations to ensure disclosures remain prominent in short-form video and social stories. |
| Australia | ASIC RG 234 (updated 9 June 2026); DDO (Design and Distribution Obligations) | RG 234 June 2026 update merged former RG 53 (past performance) into RG 234. All past performance rules now in one document. DDO Target Market Determination required for credit and investment products; audience expansion creating out-of-target reach creates DDO exposure. | Yes. ASIC enforced prominence standard in 2024-2025 actions on short-form video. |
| United States | FINRA Rule 2210 (principal pre-approval for retail communications); SEC Marketing Rule 206(4)-1 (investment advisers, effective Nov 2022) | FINRA RN 26-14 (9 July 2026) proposes risk-based supervisory standards replacing prescriptive pre-approval; comment period closes 11 Sep 2026. FINRA RN 24-09 confirmed AI-generated copy falls under Rule 2210. | Net performance must accompany gross performance at equal prominence. Cherry-picking prohibited. |
| Canada | CIRO Rule 3600 (Communications with the Public); OSFI E-23 (AI/ML model risk, effective 1 July 2027) | No confirmed 2025-2026 advertising rule changes verified. OSFI E-23 will apply to AI models in marketing decisions from July 2027. Records retained 7 years. | Performance guarantees prohibited. All materials approved by designated Supervisor before use. |
| Malaysia | BNM FTFC Section 8 (advertising); BNM RMiT (June 2023 update) for AI vendor risk | FTFC Section 8 mandates equal-prominence risk disclosures and minimum font sizes. BNM RMiT applies to AI vendors when material. No confirmed 2025-2026 advertising rule changes at time of writing. | Yes. Minimum font size for disclaimers in visual media. Compliant prominence is operationally challenging on mobile-first short-form video. |
Common misquote: MAS Notice 610
MAS Notice 610 is a statistical returns and balance-sheet reporting directive with zero advertising provisions. Practitioners who cite it as a constraint on financial advertising are wrong. Similarly, claims that a marketing agency is "MAS-exempted" are legally meaningless: marketing consultancies do not conduct regulated financial activities under the Financial Advisers Act, and the bank bears full liability for non-compliant third-party campaigns regardless of any partner's claimed exemption status.
Platform advertising constraints for financial products
Both Google and Meta impose platform-level restrictions on financial advertising that are independent of local regulatory requirements. A campaign can pass local legal review and still be rejected or restricted at the platform level.
Google Financial Advertiser Verification. Google Ads requires advertiser verification for financial products in Singapore (MAS-licence verification required since June 2022), Australia (AFSL or credit-licence verification since August 2022), the US, UK, and several other markets. The verification step adds 5-15 business days to a campaign launch. This is the operational reason that agencies without pre-verified client accounts chronically underestimate banking campaign timelines. For AI Max for Search campaigns: financial-services advertisers cannot fully opt out of auto-generated asset assembly, but feeding strong human-written assets into the account limits AI Max to selecting from a vetted pool. The existing leapbuzz post on Google Financial Advertiser Verification covers the mechanics in detail.
Meta Special Ad Category for Credit (US). For credit cards, vehicle loans, mortgages, long-term financing, and personal loans: Special Ad Category for Credit is mandatory in the US from 21 January 2025. Restrictions: targeting limited to ages 18-65+ (no age targeting), no gender exclusion, no detailed demographic interests that enable proxy discrimination, no ZIP code targeting, 15-mile minimum radius. Meta also restricted Custom Audiences built on financial indicators (income, net worth, creditworthiness) from 2 September 2025. Advantage+ Lead and Advantage+ Shopping auto-detect Special Ad Category eligibility and apply constraints automatically. The practical effect: AI audience optimisation for credit products collapses to a constrained baseline that is substantially narrower than Meta's standard Advantage+ system.
LinkedIn financial advertising. LinkedIn does not operate a financial products verification system equivalent to Google's, but financial service ads in Australia must comply with the AFSL holder requirement under ASIC's RG 234. LinkedIn's Thought Leader Ads, which serve in-feed posts from a named individual, carry lower CPMs and higher engagement than standard company-page sponsored content for financial services brands, partly because they read as personal opinion rather than institutional advertising. Compliance note: in-feed posts by named employees representing a regulated entity may constitute "communications with the public" under FINRA Rule 2210 or CIRO Rule 3600, requiring the same principal review as any other retail communication.
What banks outsource and what they keep in-house
The outsourcing boundary in bank marketing has stabilised around a consistent pattern: strategy and brand governance stay internal; execution and production go to partners. The boundary matters for compliance because everything that is outsourced carries regulatory accountability back to the bank.
Typically retained in-house: brand and corporate identity governance; compliance review and legal approval workflows (L&C); CRM and customer data management; core messaging and positioning strategy; executive and board communications; investor relations marketing.
Typically outsourced: performance marketing execution (paid search, paid social, programmatic buying); content production and thought leadership writing; social media management (execution, not strategy); influencer and ambassador programme management; PR and media relations; MarTech implementation and integration; analytics and attribution modelling.
The strategic implication for marketing partners: the outsourced functions include execution of digital campaigns, but the bank's L&C team must review and pre-approve the output. Under MAS FSG-03, this is not optional; board and senior management remain accountable for third-party campaign content. A marketing partner that cannot operate within a structured L&C review workflow, including building the compliance buffer into campaign timelines, will create friction at the most consequential point of the relationship.
The TPRM gate: the procurement track most agencies never see
Banks evaluate marketing partners through two parallel and independent procurement tracks. The marketing department track evaluates creative quality, strategic thinking, category experience, and pricing. The TPRM (Third-Party Risk Management) track, which sits in risk or operations, evaluates information security, data handling, regulatory compliance, and contract terms. These tracks run simultaneously but are not cross-informed until a vendor passes both.
A vendor who wins the CMO's creative pitch can be rejected at the TPRM gate. The creative quality is irrelevant if the vendor fails the compliance threshold. Compliance capability is pass/fail; it is not a weighted attribute that trades off against creativity scores.
The qualifications that appear most frequently in TPRM checklists for marketing partners:
- SOC 2 Type II certification for information security controls. Type II (testing controls over a period) is the minimum; Type I (point-in-time snapshot) is insufficient for most banks.
- ISO 27001 information security management system.
- GLBA Safeguards Rule compliance for US engagements (any vendor receiving non-public personal information from a US financial institution must maintain an information security programme).
- MAS Technology Risk Management guideline alignment for Singapore engagements.
- APRA CPS 234 alignment for Australian engagements (information security requirements for material service providers).
The single most common TPRM rejection point
Marketing agencies that route bank-owned data (customer personas, product strategies, yield curve data, campaign performance with identifying attributes) through public multi-tenant LLM APIs without a Zero-Data Retention agreement fail the TPRM gate. Under MAS TRM and APRA CPS 234, this constitutes a data handling breach, not merely a contractual risk. Required: a Zero-Data Retention agreement with the LLM provider, or locally hosted or enterprise-licensed LLM deployments that do not train on client data.
What banks actually evaluate in practice (beyond the formal checklist): whether the account managers can distinguish "retail client" from "accredited investor" in conversation; whether the partner will pre-reject non-compliant ideas before they reach the bank's internal L&C team, saving the marketing director from the credibility cost of a bad idea being reviewed; and whether the partner's campaign approval workflow integrates naturally with the bank's existing L&C process, including WORM-compliant archiving of all approved materials.
AI governance architecture for bank marketing
The structural challenge of AI in financial marketing in 2026 is not that AI produces bad output. It is that the volume of AI output exceeds the capacity of compliance teams to review it. AI can generate thousands of ad copy variants per day. A bank's L&C team can approve a fraction of that. The productivity gain from AI collapses unless the architecture separates what AI generates from what compliance must review.
The workable architecture separates AI's role by function:
- AI for creative imagery and audience parameters. AI-generated visuals and audience targeting parameters do not make factual claims and do not carry the regulatory risk of copy. These can be generated and deployed with a lighter review layer.
- Human-authored copy in a pre-approved matrix. All headlines, body copy variants, risk disclosures, and calls to action exist as a pre-approved, L&C-sanctioned matrix. AI selects from and arranges elements within this matrix but does not generate outside it. New claims require a new L&C approval cycle before entering the matrix.
- Automated "as of" date management for asset manager clients. One of the most operationally costly compliance failures in asset manager marketing is stale performance data. Marketing teams build campaigns in Q1 using Q4 performance data; by the time L&C approves and media buys execute, the data may be legally stale under SEC rules. A Dynamic Creative Optimisation pipeline that pulls performance data directly from the asset manager's pricing database keeps the "As of" date current without requiring template re-approval for each update cycle.
AI-generated synthetic content and disclosure obligations. EU AI Act Article 50, with obligations effective approximately August 2026, requires providers of generative AI to mark outputs in machine-readable format detectable as AI-generated where technically feasible, and deployers to inform users when interacting with AI. Banks with EU-distributed advertising face this as a direct requirement; banks without EU distribution face it as an emerging global standard. C2PA (Coalition for Content Provenance and Authenticity) v2.1 provides the technical implementation via cryptographic watermarking. Note: major social platforms currently strip C2PA metadata on upload, limiting its practical utility for social advertising channels, but it is the correct architecture for brand-controlled asset management.
AI-generated avatar and synthetic voiceover risk. AI-generated avatars or synthetic voiceovers in bank marketing create a deepfake liability risk independent of regulatory disclosure: if assets are later repurposed by threat actors for phishing campaigns, the bank's legitimate creative becomes the template for fraud. This is a reputational risk layer beyond regulatory compliance that most marketing discussions omit.
The eight-checkpoint compliance framework
The checklist below covers the minimum compliance architecture for a financial services campaign in 2026. Items marked BINARY PASS/FAIL are TPRM gates; failing them removes the vendor from consideration before creative is evaluated.
Jurisdictional content architecture
Separate content variants per market, approved by local legal. Modular geo-fenced content layers for multi-market campaigns. Regulatory basis: MAS FSG-03, ASIC RG 234 June 2026, FINRA Rule 2210, CIRO Rule 3600, BNM FTFC.
InfoSec and data residency Binary Pass/Fail
No public multi-tenant LLM APIs for bank-owned data. Evidence required: Zero-Data Retention agreement or enterprise LLM documentation; SOC 2 Type II; ISO 27001. Regulatory basis: MAS TRM, APRA CPS 234, GLBA. This is the TPRM gate with the highest weight. Failure here ends evaluation regardless of creative quality.
Metric substantiation archive (timestamped, immutable)
Timestamped screenshots of every live ad; documented source for every factual claim; retention matching local rules. FINRA: 3 years. CIRO: 7 years. Regulatory basis: FINRA Rule 2210, SEC Marketing Rule 206(4)-1, ASIC RG 234, CIRO Rule 3600.
Net performance and stale-data handling (asset manager clients)
Gross performance never presented without net at equal prominence over identical standardised time periods (1, 5, 10 years). Automated kill-switch or live data feed to prevent stale "As of" dates. Regulatory basis: SEC Marketing Rule 206(4)-1, ASIC RG 234 (incorporating former RG 53).
Risk disclosure prominence validation (all markets)
Disclaimers readable on dark mode and light mode at mobile viewport. Risk warning given equal prominence to benefit claim in the same visual frame. Tested for short-form video (15 and 30 second), static image, and stories formats. Regulatory basis: ASIC RG 234 prominence standard, MAS FSG-03 format limitations safeguard, BNM FTFC minimum font size.
Testimonial and endorsement disclosure protocol (asset manager clients)
Clear disclosure of client status and compensation for every endorser. Influencer script locked to L&C-approved version. Each testimonial person checked against SEC disqualification criteria before use. Note: many asset managers now internally ban testimonials for digital formats due to the visual burden of required disclosures. Regulatory basis: SEC Marketing Rule 206(4)-1 endorsement provisions.
AI asset lineage log
Base model, prompt, and generation timestamp logged for every AI-generated asset. EU AI Act Article 50 machine-readable marker for EU-distributed audiences. C2PA cryptographic watermark for synthetic audio and video. Regulatory basis: EU AI Act Article 50, FTC guidance on AI disclosure, MAS FEAT Principles.
Supervisor pre-approval workflow with WORM audit record
Campaign approval workflow integrates with the bank's L&C for pre-approval. Approved assets pushed to WORM-compliant archiving (Smarsh, Global Relay, or equivalent). Regulatory basis: FINRA Rule 2210, CIRO Rule 3600, SEC Marketing Rule 206(4)-1.
Two additional platform-level checks that are operationally important but sit outside the regulatory framework: pre-flight check against Meta Financial Services policy and Google Ads Financial Products policy before campaign submission (platform rejections are independent of regulatory compliance and can delay launch by 5-15 days); and verification that every ad's click-through destination contains the full jurisdiction-appropriate risk disclosures, not a generic homepage without disclosure language.