Deloitte — “Banking on Trust: AI Governance for Growth, Resilience and Scale” (14 August 2026)

Tag: S-2026-08-14-deloitte-banking-on-trust Type: report (Deloitte Trustworthy AI research, banking cut) Author(s): Deloitte (Trustworthy AI team) Date of source: 2026-08-14 Date ingested: 2026-08-28 Authority weight: medium — a Big Four firm’s own primary survey research with a disclosed methodology (135 respondents across G-SIBs/D-SIBs and other large banks); self-reported, vendor/advisory-authored (Deloitte also sells AI-assurance services), gated full PDF. Headline figures are now primary-confirmed (see Updated note below) — previously relayed only via The Financial Brand (14 Aug 2026). Raw file: S-2026-08-14-deloitte-banking-on-trust.md. Report page (primary, fetched in full 2026-09-11): deloitte.com/ap/en/perspectives/banking-on-trust.html. Coverage relayed: thefinancialbrand.com/news/artificial-intelligence-banking/your-banks-ai-ambitions-are-only-as-strong-as-your-governance-199035 (14 Aug 2026).

Updated 2026-09-11 (daily regulatory-intelligence scan — provenance upgrade, no new publication). The primary Deloitte Asia Pacific report page was fetched in full this run. It confirms every headline figure previously held only from The Financial Brand relay: 63% weekly use (doubled from 30% in 2025); 13% optimized / 87% scope to strengthen; weakest pillars organisational structure and people & skills; 5.5 vs 0.7 fully-implemented AI solutions; 72% design-vs-55% monitoring controls; 72% with <half of AI use cases in a central register; 44% agentic-lifecycle monitoring vs 61% traditional / 59% generative; 84% of consumers would switch if data mishandled; and the 10-point Governance-Index ≈ 10pp revenue-growth association, explicitly not causation. The primary page adds: authorship attributed to Deloitte Access Economics / Trustworthy AI team; only 15% of banks give regularly refreshed AI-governance training to all staff and one-third give ad hoc or no training; AI incidents for FS rose ~8x between 2022 and 2025, with more AI incidents in H1 2026 than in all of 2025; and five named recommendations (governance as a precondition for scale; visible executive accountability; AI as a transversal risk driver; move from “human in the loop” to “human on the loop”; a risk-reward culture). Authority accordingly upgraded from relay-only to primary-confirmed (still self-reported, advisory-authored, gated PDF). ⚠️ Date discrepancy surfaced (not silently resolved). The primary Deloitte page carries a publication byline of 08 July 2026; this Source page’s S-tag and source-date (2026-08-14) were set from the Financial Brand relay date, not the primary. Both dates are preserved: canonical publication ≈ 8 Jul 2026; vault ingestion driven by the 14 Aug relay. Suggested future action (not done autonomously): rename the S-tag to S-2026-07-08-deloitte-banking-on-trust to match schema §8 (S-tag = source’s own date); deferred because 15+ inbound wikilinks on AI Governance Maturity Gap, Model Risk Management and Agentic AI and others reference the current tag and would need updating in the same operation.

What it claims

Deloitte’s “Banking on Trust: AI Governance for Growth, Resilience and Scale” (14 August 2026) reports that weekly AI use among bank employees has doubled to 63% in 2026 (from 30% in 2025), while only 13% of banks have reached the report’s highest (“optimized”) level of AI-governance maturity and more than half are at “rudimentary” or “ad hoc”. Deloitte’s headline framing is that AI adoption is moving faster than the governance needed to support it, and that 87% of banks have scope to materially strengthen AI governance.

Deloitte evaluates AI governance across five dimensions: principles and policy; organisational structure; procedures and controls; people and skills; and monitoring, reporting and evaluation. The weakest pillars are human and organisational, not procedural: organisational structure is weakest (27% of banks at ad hoc), and half of banks are ad hoc or rudimentary on people and skills; procedures and controls are comparatively mature (nearly a quarter optimized). Deloitte reads this as evidence that the binding constraint is not policy but whether responsibilities are clear, decision rights understood and staff skilled to apply them — noting that heads of AI governance rated their organisations more conservatively than the broader senior-executive group, i.e. governance can “look comprehensive from the executive level while appearing much less mature to the people responsible for applying it.”

On the governance-enables-scale thesis: banks with optimized governance average 5.5 fully-implemented AI solutions across business areas versus 0.7 at the ad hoc level (Deloitte elsewhere states “nearly eight times as many”), and a 10-point increase in Deloitte’s AI Governance Index is associated with a 10-percentage-point increase in revenue growth, after controlling for AI use, workforce size, HQ location and years in operation. Deloitte explicitly cautions this is association, not causation.

On the post-deployment / lifecycle gap: 72% of banks apply mandatory governance controls during design, but only 55% do so during monitoring, and 72% of banks have less than half of their AI use cases recorded in a central register. The gap is sharpest for agentic AI: only 44% of banks report risk monitoring across the implementation lifecycle for agentic AI, versus 61% for traditional AI and 59% for generative AI. A separate consumer datum: 84% of consumers say they would switch financial providers if their data were mishandled. Deloitte also notes the financial-services industry reported more AI incidents in the first half of 2026 than in all of 2025.

Methodology (as disclosed): Deloitte surveyed 24 AI-governance leaders at G-SIBs and D-SIBs across 14 countries, plus 111 senior technology, AI and data employees at G-SIBs, D-SIBs and other large banks across 16 countries, using two purpose-built instruments — 135 respondents total.

Notable quotes

• “63% of bank employees now use AI at least weekly, while only 13% of banks have reached the report’s highest level of governance maturity.” (The Financial Brand relay, 14 Aug 2026)
• “Only 44% of banks report having risk monitoring across the implementation lifecycle for agentic AI — compared with 61% for traditional AI and 59% for generative AI.” (The Financial Brand relay)
• “A governance framework can look comprehensive from the executive level while appearing much less mature to the people responsible for applying it.” (The Financial Brand relay, paraphrasing Deloitte’s finding)

What’s speculative vs. asserted

• Asserted (Deloitte survey findings): the 63% weekly-use figure (doubled from 30%); 13% optimized / >50% rudimentary-or-ad-hoc; 87% with scope to strengthen; the five-pillar model and the organisational-structure/people-and-skills weakness; the 5.5-vs-0.7 deployment gap; the 72%-design / 55%-monitoring and 72%-under-half-registered lifecycle figures; the 44%/61%/59% agentic-vs-traditional-vs-generative monitoring split; the 84%-consumer-switching datum; the 135-respondent methodology.
• Explicitly hedged by the source: the governance–revenue relationship is presented as association, not causation (“Deloitte explicitly cautions that the analysis demonstrates an association rather than causation”) — preserve the hedge; do not read the 10-point/10pp figure as causal.
• Authority / verification caveats: self-reported survey; Deloitte is both researcher and a seller of AI-assurance services (potential framing interest); the full report is gated and figures are taken from The Financial Brand’s 14 Aug 2026 relay, not the primary PDF; sample is large global banks (G-SIB/D-SIB-weighted), not UK/EU-specific, so EU/UK read-across is directional.
• Ingesting-agent inference (not the source’s claim): the mapping to Paul’s service lines and the read that this is a competitor Big Four attesting to (and monetising) the same gap Paul’s practice targets are the wiki’s assessment.

Topics this feeds

• AI Governance Maturity Gap — a fresh, research-grade 2026 banking-specific data point squarely reinforcing the adoption-outpaces-governance thesis, and adding the sharpest lifecycle/agentic-monitoring cut yet (design 72% → monitoring 55%; agentic lifecycle monitoring only 44%).
• Model Risk Management and Agentic AI — the 44%-agentic-lifecycle-monitoring figure quantifies the agentic-oversight gap this page tracks.
• Service Line — Independent Governance Assurance — the register/monitoring gap (72% with <half of use cases registered; controls dropping from design to monitoring) is precisely the evidence-and-assurance demand this service line addresses.
• Big 4 Differentiation Frame — a Big Four incumbent publishing its own AI-governance maturity research is competitive context for the independent-specialist positioning [inference].

Open questions raised

• What is a UK/EU-only cut of the 13%-optimized and 44%-agentic-monitoring figures — do EU/UK banks track the global G-SIB/D-SIB picture or diverge (cf. Domino’s finding that Europe reports the lowest fully-integrated-governance rate)?
• How does Deloitte’s five-pillar “AI Governance Index” relate to the maturity scales used by ProSight (12% “highly developed”), McKinsey (RAI maturity 2.3) and others on the maturity-gap page — are these measuring the same construct differently?
• Does the “72% have <half of AI use cases in a central register” figure reconcile with the AI-inventory expectations emerging under the EU AI Act and firms’ own model inventories?