Shape the AI Bank before it shapes you

We are living through the fastest technology transition in the history of finance. Reasoning models now outperform expert humans on medical and legal licensing exams. Multimodal AI reads documents, interprets charts and listens to calls in real time. Agentic AI systems execute multi-step tasks autonomously — booking, transacting, reconciling — without a human in the loop. And the cost of intelligence has fallen by more than 99% in three years.

For banks, this is not a background trend. It is a structural disruption arriving simultaneously on every front: customer experience, credit decisioning, fraud detection, compliance, treasury, wealth management and the operating model itself. The question is no longer whether AI will transform your institution — it is whether your leadership team will shape that transformation or be shaped by it.

This two-day Bootcamp is built around three convictions: that understanding must be current (the AI landscape of six months ago is already obsolete); that strategy must be hands-on (knowing about AI is not the same as knowing how to use it); and that decisions must be made. By the end of Day Two, every participant will have built a draft AI Bank concept — stress-tested by peers and ready to take to their board.

Get Your Brochure

For Past Event Photos Click Here

Skills to be gained

Shape executive strategies for future banking

Lead technology-driven transformation

Strengthen decision-making and governance

Design customer-first product innovations

Manage data, ethics and trust

Drive organisational change

Expand peer networks across the region

Areas to be covered

Banking in an intelligent economy

Evolving technology architecture

Data governance and cybersecurity

Customer engagement and new products

Leadership and talent strategies

Fairness, ethics and trust

Regulatory models and public policy

Strategic Briefing · Banking AI Bootcamp 2026

The Landscape Has Changed: What Bankers Must Know Now

The five developments below define the strategic context for this Bootcamp. They are not predictions — they are live deployments reshaping competitive advantage in banking right now.

01
Foundational banking models

The new competitive moat

Banks are moving from fragmented AI use cases to unified models trained on proprietary data. Competitive advantage is no longer the algorithm. It is the depth and structure of your data.

Banks are moving from fragmented AI use cases to unified models trained on proprietary data. Competitive advantage is no longer the algorithm. It is the depth and structure of your data.

Read More
02
Agentic core banking architecture

Moving beyond applications

AI agents are beginning to orchestrate banking processes across systems. Over time, business logic shifts out of applications and into an agent layer, changing how banks are built.

AI agents are beginning to orchestrate banking processes across systems. Over time, business logic shifts out of applications and into an agent layer, changing how banks are built.

Read More
03
The end of the interface

AI becomes the front end

AI is replacing traditional interfaces across enterprise systems. For banks, this changes how customers interact, how workflows operate, and how vendors integrate.

AI is replacing traditional interfaces across enterprise systems. For banks, this changes how customers interact, how workflows operate, and how vendors integrate.

Read More
04
Agent-to-agent commerce

A near-term reality

Transactions will increasingly be negotiated and executed by AI agents. Customer journeys, pricing models and authentication frameworks will need to be redesigned.

Transactions will increasingly be negotiated and executed by AI agents. Customer journeys, pricing models and authentication frameworks will need to be redesigned.

Read More
05
The regulatory reckoning

From principles to enforcement

AI governance is now being enforced through regulation. Boards must address model risk, explainability and accountability across all AI-driven functions.

AI governance is now being enforced through regulation. Boards must address model risk, explainability and accountability across all AI-driven functions.

Read More
Foundational Banking Models: The New Competitive Moat

The most consequential development in banking AI is not a chatbot or a copilot — it is the emergence of domain-specific foundational models trained on proprietary banking event sequences. Revolut's PRAGMA (April 2026) is the most advanced public example: a single pre-trained backbone that simultaneously outperforms task-specific models across credit scoring, fraud detection, lifetime value prediction, product recommendation and more. Stripe, Mastercard and Visa have all announced comparable models. The moat is no longer the algorithm — it is proprietary data at scale. Any bank sitting on comparable transaction volumes could build its own version. Most are not.

Read More
The Agentic Core Banking Architecture: Moving Beyond Applications

Today's bank technology estate is built around vertical application silos — lending, treasury, payments, CRM — each a walled garden connected by brittle integration layers. A rigorous new practitioner framework describes the three-phase migration path out of this architecture: Phase 1, agents orchestrate across existing systems via API; Phase 2, new products are built as pure agentic processes and business logic begins migrating out of applications; Phase 3, applications flatten to data access layers and the concept of a 'banking application' becomes obsolete. The critical implication: data governance investment must precede architectural migration, not follow it.

Read More
MCP and the Death of the Interface

Anthropic's Model Context Protocol — now backed by OpenAI, Google, Microsoft and AWS under the Linux Foundation — has fundamentally changed how enterprise software is built. Major platforms including Slack, Figma and Asana now embed their interfaces directly inside AI chat windows. The app becomes a backend; the interface is owned by the AI. For banks, every customer touchpoint, every workflow tool and every vendor relationship must be re-evaluated against an agentic architecture.

Read More
Agent-to-Agent Commerce: The 12-Month Horizon

When a customer's AI agent negotiates with a bank's AI agent to arrange a mortgage, the entire customer journey, pricing logic, authentication framework and relationship model must be redesigned. Early agent-to-agent payment rails are already live on Stripe's Agent Commerce Protocol. Banks that have not begun designing for this world are already behind — and foundational models make the gap wider still.

Read More
The Regulatory Reckoning

The EU AI Act is now in force for high-risk applications including credit scoring, KYC and fraud systems. Bank Negara Malaysia and MAS Singapore have both issued updated guidance on model risk, algorithmic fairness and AI governance. Boards must understand what this means for model documentation, explainability requirements and third-party vendor oversight — especially when a single foundational model underlies multiple regulated functions simultaneously.

Read More
Pre-Bootcamp Preparation
  • Pre-reading pack distributed three weeks before the event—including practitioner perspectives on foundation models, agentic banking architecture, current African AI and financial-sector regulatory developments, and selected chapters from Building the AI Bank by Emmanuel Daniel.
  • Pre-event survey: a 15-minute diagnostic to calibrate session depth and identify the specific architecture challenges and competitive pressures most relevant to the group
  • Tech setup guide: instructions for accessing Claude and the lab session tools, with a 30-minute self-paced orientation for participants new to frontier AI tools
  • Optional pre-bootcamp webinar: a 60-minute orientation session one week before the event
Post-Bootcamp Deliverables
  • Proceedings Report summarising key insights, areas of consensus and the AI Bank concepts from the Capstone — shared within 48 hours
  • All presentation materials and lab guides, including agent design frameworks, foundational model assessment templates and Capstone project materials
  • Each participant's AI Bank concept document and agent design documentation, formatted for internal handover
  • Delegate contact directory and LinkedIn group for ongoing peer exchange
  • One-year membership to The Banking Academy's Future Bank Working Group — quarterly briefings, webinars and research
  • Optional: 60-minute follow-up strategy session with a Banking Academy associate within 60 days

Who should attend

Chairmen and board members of traditional and digital banks

CEOs and senior management

Heads of strategic planning, innovation and business development

Heads of AI, data, analytics, cybersecurity, channels and distribution

Senior managers driving AI adoption across banking functions

Regulators and policymakers shaping AI governance in financial services

AI Banking Bootcamp · Programme Format

Types of Sessions

Expert Briefing

A structured intelligence briefing on live deployments and current developments, followed by facilitated discussion.

Hands-On Lab

Participants work directly with AI tools on real banking scenarios. Laptops required. Output produced during the session.

Architecture Workshop

Small groups interrogate a live architectural question — applying the three-phase core banking migration model or the PRAGMA paradigm to their own institution.

Structured Dialogue

SmallFacilitated peer discussion using real-world scenarios and challenge cards.

Build Sprint

Participants use AI to construct a real deliverable in real time — a workflow, agent design, policy draft or strategy document.

Capstone Project

Groups design a functioning AI Bank concept, using AI to produce their presentation deck or prototype live — presented to peers at the close of Day Two.

The Programme

Building the AI Bank Bootcamp Africa 2026 | Pullman, Cape Town, South Africa

6 October 2026

18:00 – 20:00

Welcome Networking Cocktail — Introductions and Expectations

Networking

An informal evening to meet fellow participants and the faculty before the programme begins.

  • Each delegate introduces themselves: who you are, your institution and your role in its AI journey
  • Each delegate states one expectation: the question you most want answered, or the thing you most want to take home
  • The faculty introduce themselves and the shape of the two days

7 October 2026

08:00 – 09:00

Registration | Device check for the hands-on sessions, Wi-Fi credentials provided

08:30 – 09:15

Opening Session — The State of AI in Banking: Latest Developments in the Industry

Big Picture

A rapid-fire intelligence briefing grounded in what is actually deployed, not what is promised. We move fast, stay concrete, and start from the room's own answers.

  • The AI Maturity Index. Each delegate scores their own institution on a one-page checklist: retrieval-augmented generation in use; MCP connections in place; human skills identified and tagged; agentic experiments running; structures evolving; outcomes measured; guardrails in place. The room's scores are shown on screen and set the baseline for the two days.
  • From digital bank to AI-native bank: why a bank that adds AI tools to a digital estate is not an AI bank, and what it means when the conversation is the product
  • The agentic turn: an agent is AI that understands, reasons, decides and acts. Which of the four does your bank allow today? Back-office agents are now handling the exceptions that rules could not, at global institutions and at South African banks
  • When your customer is an AI bot: agentic commerce is live when card networks let AI agents pay, and AI assistants reach into the customer's bank account
  • Frontier AI as a security event: models that find software vulnerabilities faster than banks can patch
  • The twelve-month horizon: the decisions your board needs to make today.

Facilitated by Emmanuel Daniel

09:15 – 10:45

Hands-On Session 1 — Lay the Land and Build It: From Blank Screen to a Running Agent

Hands-On Lab

This session brings every participant to the same starting point: able to design and run an agent, a basic skill that will soon be assumed before any discussion begins. Most senior banking audiences have watched AI being demonstrated. Far fewer have configured an agent, watched it fail, and had to decide whether they would put their name to it. By 10:45 every table has a working AI agent doing a real piece of back-office banking work. It will be rough and it will get things wrong, and that is the point: its failures are the material for the rest of the day.

  • What an agent actually is, in the terms you will configure in the next hour: the model, its tools, skills, memory, knowledge and delegated sub-agents
  • The sorting exercise: ten real back-office tasks. Your table decides which are chatbot work, which are ordinary software, and which need an agent. It hands you the most useful question you can put to a vendor next quarter
  • Worked example from the front: a third-party and outsourcing risk agent, built live, so you see every building block before you touch one
  • Choose your case: financial crime alert investigation; suspense and payment exception resolution; card disputes; credit early warning; end-to-end customer due diligence review
  • Build version 0.1: identity, objectives and constraints; the tools and data it can reach; what it needs to read. Technical delegates drive, business delegates interrogate

Output: agent v0.1 running one task end to end; your table's working document (completed Agent Opportunity Scorecard) started; the first two slides of your table's deck generated by AI; one agreed real banking use case; selected agent pattern.

Facilitated by Tavonga Muchuchuti

10:45 – 11:00

Coffee & Discussion Break

11:00 – 11:45

Session 2 — Agentic AI in Banking: The Hard Problems That Boards Are Not Asking Yet

Tutorial

A briefing on the governance, liability and architecture questions practitioners are wrestling with now.

  • The future of agentic AI, as a checklist: AI understands, AI reasons, AI decides, AI acts. Where in that chain does your bank stop the machine today, and where should it?
  • Authentication and identity: when an AI agent presents a claim on behalf of your customer, how do you verify it, and who is liable when it costs money?
  • Software versus agents: software does what it was written to do; an agent decides what to do next. Which of your processes should never be given that discretion, and what changes when an agent is placed inside existing software?
  • Designing an MCP: the Model Context Protocol is the contract between the agent and your systems. When the MCP says "I do not have enough information to act on this decision", who designs that refusal, and what happens next?
  • Agentic commerce: agent-led payments inside consumer AI assistants and machine-to-machine payment rails are now live. What does a bank need to be reachable by an agent?
  • Orchestration architecture: single-agent versus manager-agent patterns, and why they fail differently. Why the manager-agent pattern is becoming the dominant enterprise architecture and what it means for bank technology design.
  • Agent memory and context: persistent memory, retrieval-augmented generation and the privacy implications for banking. When your customer's AI knows more about them than your CRM does.
  • The MCP ecosystem in banking: which institutions are building MCP-compatible infrastructure today, what tools are live, and what the 'headless bank' architecture looks like in practice
  • Guardrails and governance: the technical and organisational mechanisms that actually work at scale — not principles but operational implementations

Discussion: map the three banking processes in your institution most exposed to agent disruption in the next 12 months — and three where you could deploy agents to gain competitive advantage.

Facilitated by Axel Winter

11:45 – 12:15

Session 3 — Agents and Humans: The New Hybrid Workforce

Dialogue

Exercise: each table allocates twelve pieces of work between people and agents, then names the human who supervises the agent and the limits on its permissions, spending and data access. Tables compare answers before the briefing begins.

AI agents take on work that used to belong to analysts, operations staff and relationship teams. Banks therefore face a new management task: running a workforce in which some of the workers are not human. This session looks at five questions:

  • Collaboration: how to divide work between people and agents so each does what it does best
  • Leading agentic workers: how to assign tasks, delegate authority, set escalation paths and keep clear accountability for agents
  • Quality control: how to monitor, sample, review and measure agent output, rather than relying on spot checks after something goes wrong
  • Guardrails: which limits on permissions, spending, data access and customer interaction keep agents within the bank's risk appetite
  • The junior-banker problem: banks are cutting entry-level intakes as agents take on analyst work. If agents do the entry-level work, who becomes your senior credit officer in 2035?

Participants will leave with a practical view of what changes in supervision, performance management and governance when the team includes agents.

What do we tell our board?: each delegate writes one sentence on the hybrid workforce for their own board; three are read aloud and pinned to the Board Wall.

Facilitated by Prof. Juergen Rahmel

12:15 – 13:15

Lunch — Facilitated Peer Tables: 'The industry development that most changed how I think about our institution'

13:15 – 14:15

Session 4 — Foundational Models in Banking: Two Roads to the Data Moat

Tutorial

This session examines the most consequential architectural development in financial AI — the emergence of institution-wide foundational models trained on a bank's own proprietary event data: one shared backbone that replaces dozens of siloed models. We will work through the strategic and data governance implications for every bank in this room. Two cases show two different roads; the second matters most for this room.

  • Revolut's PRAGMA: a single backbone trained on about 40 billion banking events from 25 million users, scaling from 10M to 1B parameters and fine-tuned on 1–2% of its parameters. A single shared backbone that replaces siloed task-specific models, outperforming task-specific models across credit scoring, fraud, lifetime value predictions, communication engagement, product recommendation and recurrent transaction detection simultaneously.
  • Nubank's nuFormer: the emerging-market road. A proprietary foundation model live in Brazil's largest credit segment and expanding across Latin America, built to extend credit safely to more people, with an AI private banker already serving millions
  • The recipe is now public. The moat is the data, not the algorithm. Any bank with comparable transaction volumes could build its own version. Most are not.
  • The data governance question: unified multi-source event data across products, channels and time, and why most bank data estates were not built for it
  • The architecture in plain terms: key-value-time tokenisation of heterogeneous financial events (transactions, navigation, trading, communications), two-encoder branch design fusing profile state and event history, masked modelling pre-training. LoRA fine-tuning on 1–2% of parameters outperforms full retraining from scratch. Production inference on 200+ NVIDIA H100 GPUs, 80% of customer support resolved without human escalation.
  • The competitive implications: Stripe, Mastercard and Visa have all announced comparable models. The convergence means the moat is no longer the algorithm — it is the proprietary event-scale data. Any institution with sufficient transaction volumes could build a version of this. Most are not thinking about it. The institutions that begin structuring their data for foundational model training now will have a structural advantage that compounds over years.
  • What this means for every downstream banking function: when credit scoring, fraud, LTV, communication engagement and product recommendation all run from the same foundational representation of each customer, the economics of AI deployment change fundamentally. Update one backbone and every downstream task improves simultaneously.
  • The data governance question: to train a PRAGMA-equivalent, you need multi-source event data unified across products, channels and time. Getting all data 'right' — not just analytics data but operational and unstructured data — is the prerequisite. Practitioners building this at scale are clear: most banks' current data architecture was not built for this, and the investment must start now.

Exercises

  • Debate: "Our bank should build its own transaction foundation model within 24 months." Two volunteers per side, three minutes each, a room vote before and after
  • Assignment: assess your institution's current data estate against the requirements for a foundational model programme in a data scorecard — what do you have, what is missing, and what is the 18-month path to close the gap?

Facilitated by Axel Winter and Prof. Juergen Rahmel

14:15 – 15:00

Session 5 — What Does a Fully AI-Native Core Banking System Look Like? Dismantling the Application Silo Model

Design

Most discussions of AI in banking focus on use cases: fraud detection here, customer service there. This session addresses the harder question: what does a bank's entire technology architecture look like when it is rebuilt for an agentic, foundation-model world? The answer, emerging from practitioners at the most advanced AI banking programmes, dismantles assumptions that most boards and technology teams still take for granted.

  • The diagnosis: vertical application silos, each with its own interface, logic and database, stitched together by integration layers that consume engineering effort and cause a disproportionate share of incidents
  • Phase 1, agentic orchestration: agents call existing applications through their APIs. Phase 2, logic migration: new products are built as agentic processes. Phase 3, application thinning: applications flatten into data access layers under an agentic processing tier
  • The data governance consequence: the shift from Phase 2 to Phase 3 requires all data, not just reporting data but operational and unstructured data, to be governed correctly. Banks already building agentic architectures are unanimous: data governance investment must precede the migration, not follow it
  • Your agent is a Phase 1 agent: it calls mocked systems today, and replacing a mock with a live core banking interface is a configuration change, not a redesign. Understanding why is itself worth taking home

Architecture workshop: apply the three-phase model to your institution. Where are you today? What is the primary constraint holding you in Phase 1? What would Phase 3 look like for your specific product and regulatory environment?

Facilitated by Axel Winter

15:00 – 15:30

Session 6 — Creating a Decisioning Architecture: Who Decides, Who Supervises, Who Can Stop It

Design

A digital bank adds AI to review applications faster. An AI-native bank designs credit, fraud and payments assuming machines will act first and humans will supervise. Banks that treat AI as a programme ask how many use cases they have deployed. AI banks ask how many material decisions are now machine-supported, and how they are governed. This session is about designing that answer before scale.

Exercise: list the decisions your table's agent makes, from the first lookup to the last action. For each one: is it reversible, what is at stake, how often is it made, and can you tell afterwards whether it was right? Those four tests decide everything that follows.

  • The real AI stack of a bank is functional, not technical: Data, Inference, Control, Execution. The control layer decides what AI is allowed to do, when it must ask for help and when it must stop; the decision boundary is the line between acting alone and calling a human
  • Five patterns for placing the human: human-led with AI as challenger; AI-supported, where the human decides on richer inputs; active approval, where AI recommends with its reasoning exposed; AI with exceptions, where the machine decides the routine within a mandate and escalates the rest; AI with oversight, where the machine runs continuously and humans watch for drift
  • Autonomy and agency are different dials: how much a human is in the loop, and how far the system can reach and whether its actions can be undone. Five design moves: fixed checkpoints, agent-initiated escalation, delegation to specialist agents, provisioning and fencing of tools, and staging writes so that a commit can be rolled back
  • Where the MCP says "I do not have enough information to act on this decision": escalation designed in, not bolted on
  • Three tests for every human sign-off: has the reviewer ever rejected it (if approval runs above 95% the checkpoint is theatre); what does an error cost; does the review teach the system anything? Give every checkpoint a sunset date
  • Decision governance that works: a taxonomy of decision types, allocation reviewed on a known cadence, monitoring built for each pattern's failure mode, every decision captured with its inputs and reasoning, and explicit links to model risk, enterprise risk and the board

What do we tell our board? he first material decision in your bank that should move to "AI with exceptions", and who signs the mandate.

Facilitated by Emmanuel Daniel and Prof. Juergen Rahmel

15:30 – 15:50

Afternoon Break

15:50 – 17:00

Hands-On Session 2 — Build the Agent and Make It Trustworthy

Hands-On Lab

This second hands-on session is the most valuable for this room, because everyone in it is accountable for something. Business delegates now drive and technical delegates interrogate.

  • Fix what broke in Hands-On Session 1
  • Apply your decisioning architecture from Session 6: assign each of the agent's decisions to a pattern, from AI with oversight to active approval
  • Skills: the procedures your bank has already written down somewhere, taught to the agent once rather than repeated
  • Memory: the standing rules that should not need restating
  • Delegation: sub-agents under an orchestrator, where "AI workforce" stops being a phrase and becomes something you have configured
  • Design the MCP: define the tools and data your agent may call, and build the path where it says "I do not have enough information to act on this decision", stops, and escalates to a named human
  • Guardrails: what the agent must never do however it is asked, where a human signs, and whether its logs would satisfy an auditor
  • Run the agent on the path it was designed for and record the result in your working document

Output: agent v0.5 with an audit trail; slides three to six of your table's deck drafted by AI from the working document.

Facilitated by Tavonga Muchuchuti

17:00 – 17:30

Day One Close — Theme Desk: You Propose, the Room Responds

Dialogue

Each table in three sentences: what our agent does, what broke, what we fixed.

Facilitated by Tavonga Muchuchuti and Emmanuel Daniel, with delegates

19:00

Hosted Networking Dinner


8 October 2026

07:30 – 08:30

Breakfast

09:00 – 09:30

Session 7 — Where AI Lives: Structure and Accountability

Dialogue

Exercise: your AI agent declined 400 SME loans overnight. Which executive answers for it? Each table names the person in its own bank, then places its institution on the spectrum from central AI lab to embedded squads to hybrid. The two ends explain their choice.

  • Three operating models: the central AI lab, squads embedded in the business, and the hybrid of a centre of excellence plus embedded teams, now the dominant pattern in regulated institutions
  • The new roles, stripped down: what a Chief AI Officer, an AI Product Manager and an AI Risk Officer actually decide, and whether your CRO, CDO and COO can take those accountabilities on

What do we tell our board?: one sentence each on who should own AI in your bank; three read aloud.

Facilitated by Prof. Juergen Rahmel

09:30 – 10:00

Session 8 — The Last Mile and Re-energising the People: Why Pilots Don't Scale, and Reskilling

Dialogue

Exercise: bring-your-own-pilot. One delegate per table describes a live AI pilot from their own bank in two minutes, and the table runs a pre-mortem: the three most likely reasons it dies before production.

  • Four barriers, one banking example each: incentives, governance that says "no" by default, no owner after the pilot team disbands, and frontline resistance
  • Reskilling and re-energising the people: broad AI literacy for everyone versus deep role-based skills for a few; identifying and tagging the human skills the bank must keep; what DBS's AI literacy programme and Standard Bank's enterprise-wide adoption have learned; and how to give people who fear replacement a reason to lead the change
  • Each table names the single organisational intervention that would move its bank up one stage, and presents it in 60 seconds

What do we tell our board?: one sentence each on why our pilots stall; three read aloud.

Facilitated by Prof. Juergen Rahmel

10:00 – 10:30

Session 9 — Ethics and Bias: It Is in the Plumbing

Dialogue

Exercise: the Bias Hunt. Each table inspects its own agent for one of four entry points for bias (credit scoring, collections prioritisation, fraud flags, marketing eligibility) and writes the fix into its guardrails before the briefing.

  • Ethics is not a layer added at the end. It sits in the data, the labels, the objective function and the feedback loops
  • The foundation-model problem: if a PRAGMA- or nuFormer-style backbone learns historical discrimination, retraining downstream models does not remove it. The fix has to happen at the backbone
  • Speed without due process: Phase 2 and Phase 3 architectures need escalation points, reversibility and named owners built in from the start

What do we tell our board?: one sentence each on the bias your board is not asking about; three read aloud.

Facilitated by Prof. Juergen Rahmel

10:30 – 11:00

Coffee Break

11:00 – 11:30

Session 10 — Good Governance: Trust by Design – Rules, Frameworks and the Manipulation Line

Dialogue

Exercise: Regulator's Question Time. Two delegates answer for their own bank on a live question; the regulator is played by a delegate from a different jurisdiction, and the room scores the answers.

  • What good governance means: good governance does not mean things do not go wrong. It means the bank knows when they have, who answers, how the decision is reversed, and what it learns. A governance framework that promises nothing will fail is the one that fails first
  • EU AI Act, what compliance looks like in practice: what "high-risk" obligations mean day to day when one backbone model serves several regulated functions (documentation, logging, human oversight, conformity assessment). The dates matter: the Digital Omnibus moved Annex III high-risk obligations, which cover creditworthiness assessment, from 2 August 2026 to 2 December 2027. It entered into force on 27 July 2026.
  • Singapore and Malaysia: MAS's Guidelines on AI Risk Management are still described as proposed, alongside a practical MindForge operationalisation handbook. In Malaysia, the current reference is the industry-led AI Governance Framework, which BNM publicises but did not issue itself.
  • DBS's PURE framework (Purposeful, Unsurprising, Respectful, Explainable) as the practitioner test; vote "helpful" or "manipulative" on three hyperpersonalisation scenarios
  • South Africa: the FSCA's principles-based approach, its alignment with the Financial Stability Board's AI principles due this month, and the joint FSCA, Prudential Authority and SARB consultation that follows
  • Botswana, Malawi and the DRC: where AI-specific rules are silent, what supervisors will still ask under model risk, outsourcing and consumer protection
  • The manipulation line: hyperpersonalisation at scale. Two or three short scenarios; participants vote "helpful" or "manipulative" on each before discussing. "Unsurprising", from PURE, is the practical test.

What do we tell our board?: participants write one "plumbing fix" each to take back to their institution; three read aloud.

Facilitated by Prof. Juergen Rahmel

11:30 – 12:30

Session 11 — Cybersecurity in the AI Age: New Threats, New Defences

Tutorial

AI has fundamentally changed the threat landscape. The attack surface is now cognitive as well as technical, and foundation models and agentic architectures create vectors that conventional security frameworks were not designed for.

  • Frontier models that find software vulnerabilities at scale: the restricted release of a vulnerability-finding model this year prompted warnings to bank chiefs from the US Treasury and Federal Reserve and emergency meetings at several central banks. Patching measured in quarters is no longer a defence
  • When identity collapses: zero trust where the requester may be an AI agent, a deepfake or a real customer. How do you authenticate intent, not just identity?
  • Protocol risk is operational risk: prompt injection, tool poisoning and context manipulation in MCP and agent frameworks, and the tail risk of one backbone model failing across credit, fraud and compliance at once
  • Post-quantum cryptography: NIST standards are published and migration planning cannot wait for 2028
  • Tabletop: 16:55 on a Friday, a voice-cloned CFO instructs your payments agent. Each table walks through what its own agent would do, minute by minute

Facilitated by Axel Winter

12:30 – 13:30

Working Lunch — in capstone tables

13:30 – 14:30

Role Play — The Board Meeting: Approving an AI Strategy

Board Simulation

Delegates are divided into two or three boards, each governing a different bank in the same fictional developed market: a digital-only bank, a mid-sized bank and, if numbers allow, the largest bank in the country. Each board receives the same paper from management asking it to approve a five-year AI strategy, but each bank starts from a very different position, and all of them compete for the same customers.

  • Every delegate takes a board seat with a public position and a confidential brief: Chair, CEO, CFO, Chief Risk Officer, and independent directors for technology, audit, and people
  • The decisions: whether to own a model; build, buy or partner; how to structure for the skills the bank has; who owns AI and which committee oversees it; what timeline to set and what would make the board stop; and what happens to the people
  • During their deliberations, the boards must respond to a series of structured interventions from the faculty, in role as the regulator, the investor and the customer, which test whether the strategy holds
  • Each Chair reads out the board's resolution; the faculty judge which bank gains ground by 2028, and why

Output: one signed board resolution per bank, carried into the Capstone and the Showcase.

What do we tell our board?: one sentence each on the question from this meeting that your own board has not yet answered.

Facilitated by Emmanuel Daniel, with all faculty in role

14:30 – 15:00

Afternoon Break — decks submitted and loaded

15:00 – 16:30

Hands-On Session 3 — Capstone: Prove It and Take It Home

Capstone

This is the session the Bootcamp has been building towards. Building an agent is not the hard part. Getting predictable behaviour under conditions you did not anticipate is. This session proves your agent, turns it into a business case and finishes the deck you will present.

  • Run it three ways: the path it was designed for, an edge case it was not quite designed for, and a deliberate attempt to break it by a table from another jurisdiction. Something fails, and you fix it
  • The business: what success looks like, what gets counted, which signal would show it going wrong, who owns it, and what it takes to reach production
  • Assemble the deck: AI generates your table's final ten-slide PowerPoint from the completed working document; the table cuts, edits and gives every member a speaking part
  • Finish your personal board pack: three slides on your own bank, drafted with AI
  • The three parallel realities test: your AI Bank must hold together across the technological reality (what can you actually build?), the competitive reality (who does this beat?) and the institutional reality (can a regulated bank operate this?)

Output: your table's PowerPoint deck and working agent, and your personal board pack, submitted by 16:20.

Facilitated by Tavonga Muchuchuti and all faculty in role

16:30 – 17:00

Capstone Presentation and Discussion — AI Bank Showcase: Present, Challenge, Commit

Showcase

The final session is where the learning becomes accountability. Each table presents its own AI-built PowerPoint deck and runs its agent live. No pre-prepared slides. What you built is what you present, and every member of the table speaks.

  • Eight minutes per table: the problem and its cost, the AI Bank thesis, what you built, the guardrails, how it performed under test, and how you would measure and own it
  • Two rounds of structured peer challenge, the same two every time and announced on Day One: "What in this would collapse under regulatory scrutiny?" and "What assumption have you not tested?" The challenging table rotates and comes from a different jurisdiction where possible
  • Strong answers are specific: "that is fair, and here is how we would test it within 90 days" beats a defence
  • Adjudication by a panel of senior delegates and faculty, and a closing synthesis across every table's deck

Facilitated by all faculty

17:00

Feedback and Conclusion — Commit

Close

17:00

End of Programme — Certificate presentation and group photograph

Anchored by a highly qualified team of experts

The programme is anchored and delivered by some of the most experienced professionals in the industry.

Emmanuel Daniel
Emmanuel Daniel
Founder, TAB Global
Axel-Winter
Axel Winter
AI Operating Model and
Enterprise Architecture Leader
Former Global Head, Standard Chartered
Barry Katz
Juergen Rahmel
Banking AI Research and
Model Risk Leader
Lecturer, University of Hong Kong
Tavonga-Muchuchuti
Tavonga Muchuchuti
AI Policy and Financial Innovation Leader
Board Member, Africa Fintech Network

The Investment

With Accommodation

US$ 5,800

  • Full access to Building the AI Bank Bootcamp Africa 2026
    (7 - 8 October 2026, Cape Town, South Africa).
  • Programme slides, reading materials and videos.
  • 3-night accommodation in Cape Town, South Africa
    (6 - 8 October 2026).
  • All program-related breakfasts and lunches.
  • Certificate of Participation
  • One year membership to Future Bank Working Group

*Excludes airfare, visa fees, airport transfers and dinners

Without Accommodation

US$ 5,000

  • Full access to Building the AI Bank Bootcamp Africa 2026
    (7 - 8 October 2026, Cape Town, South Africa).
  • Programme slides, reading materials and videos.
  • All program-related breakfasts and lunches.
  • Certificate of Participation
  • One year membership to Future Bank Working Group

*Excludes airfare, visa fees, airport transfers and dinners

Have a Query? Get in Touch with Us Today!

4656

A PHP Error was encountered

Severity: Notice

Message: Undefined variable: captcha_sum

Filename: template/building-the-ai-bank-bootcamps-mea-2026.php

Line Number: 1274

Backtrace:

File: /var/www/thebankingacademy.com/html/tbabetalive/tbabetalive/application/views/template/building-the-ai-bank-bootcamps-mea-2026.php
Line: 1274
Function: _error_handler

File: /var/www/thebankingacademy.com/html/tbabetalive/tbabetalive/application/controllers/Front.php
Line: 437
Function: view

File: /var/www/thebankingacademy.com/html/tbabetalive/tbabetalive/index.php
Line: 317
Function: require_once

">

Delegate Registration

Haven’t Booked Your Seat Yet?

Get In Touch

For delegate registration, click here

Have a Query? Get in Touch with Us Today!

9471

Delegate Registration

Chat with us WhatsApp