Singapore’s MAS AI training pledge is a commitment by a pioneer batch of 23 banks, insurers and asset managers, convened by the Monetary Authority of Singapore (MAS), to train 80,000 local employees in AI skills by 2028. Announced on 24 September 2026, it names specific functions, wealth advisory, compliance and operations, rather than offering generic AI literacy, and it pairs training with formal job redesign work.
For L&D leaders outside Singapore and outside finance, the pledge is worth studying closely. It is the largest sector-specific, government-convened AI upskilling commitment published so far, and it gives a concrete answer to a question most regulated industries are still avoiding: what do you actually do, operationally, when a regulator asks an entire sector to get its workforce AI-ready on a deadline.
This piece breaks down what was pledged, how it differs from the smaller bank-only commitment that preceded it, and what a training team in any regulated industry, healthcare, aviation, government, energy, can borrow from the approach: role-differentiated curricula, AI-oversight assessment, and a way to evidence compliance-relevant AI training when a regulator asks for proof.
What exactly did MAS and the 23 financial institutions commit to?
MAS convened 23 banks, insurers and asset managers into what the Institute of Banking and Finance (IBF) calls the AI Workforce Co-Lab, with a pledge to train 80,000 local employees in AI skills by 2028. The commitment covers leadership roles, wealth management, and operations staff, and it was announced alongside a companion Job Redesign Playbook for Financial Services.
Deputy Prime Minister Gan Kim Yong, who chairs MAS, announced the pledge on 24 September 2026 at the Institute of Banking and Finance (IBF) Distinction Evening, a banking and finance sector dinner in Singapore. He framed the goal in workforce terms rather than technology terms: “We want AI to translate into better careers for Singaporeans and a more productive workforce.”
MAS Managing Director Chia Der Jiun, who also chairs IBF, made the connection between training and role change explicit: “Equipping workers with new emerging skills is only part of the equation. We also need to support workers as jobs and tasks evolve.”
The Job Redesign Playbook was developed by IBF together with the Institute for Human Resource Professionals (IHRP) and the Skills and Workforce Development Agency, according to IBF’s own announcement of the AI Workforce Co-Lab launch. That pairing, a training body, an HR professional body and a workforce agency co-authoring one document, is itself a signal: MAS is treating AI upskilling as a job-architecture problem, not a course catalogue problem.
Financial services account for roughly 14% of Singapore’s economy and employ around 200,000 people, based on figures reported alongside the announcement. The 23 institutions in the pioneer batch are described in reporting only by category, banks, insurers and asset managers, not by a complete named list; MAS and IBF have not published the full roster publicly at the time of writing, so this article does not name individual institutions.
How does the 80,000-employee pledge compare with the earlier 35,000-employee bank commitment?
The new pledge is roughly double the headcount of the prior one and more than triple the number of employers involved. Where the 2025 commitment came from Singapore’s three local banks alone, the 2026 pledge spans an entire sector, banks, insurers and asset managers, with the same 2028 completion horizon carried forward rather than extended.
Singapore’s three local banks committed roughly a year earlier, around September 2025, to train all of their Singapore-based employees, a group reported at around 35,000, in AI skills. That commitment ran alongside AI-focused bootcamp programmes covering hands-on use of generative and agentic tools inside individual banks, as Bloomberg’s reporting on one bank’s AI bootcamp describes.
| Dimension | 2025 bank commitment | 2026 sector-wide pledge |
|---|---|---|
| Employers involved | 3 local banks | 23 banks, insurers and asset managers |
| Employees covered | ~35,000 | ~80,000 |
| Completion deadline | Not fixed at a single sector-wide date | 2028 |
| Sector scope | Banking only | Banking, insurance, asset management |
| Announced | Around September 2025 | 24 September 2026 |
| Companion output | Individual bank AI bootcamps | Job Redesign Playbook for Financial Services |
The scope change matters more than the headcount change. Insurers and asset managers have different regulatory obligations, different client-facing roles and different existing training infrastructure than retail and commercial banks. Coordinating one pledge across all three sub-sectors is a harder operational problem than three banks each running their own programme.
Which job functions does the training actually target?
The pledge names three groups explicitly: leaders, wealth managers, and operations staff. That is narrower and more specific than a blanket “AI literacy for everyone” goal, and it points straight at the functions where AI tools are already changing daily work fastest.
Leadership training, in this context, is less about tool use and more about oversight: knowing when to trust an AI-generated recommendation, when to escalate, and how to hold a team accountable for outcomes an algorithm helped produce. Wealth management training addresses a genuinely different problem, advisers using AI to draft recommendations or summarise portfolios while remaining personally liable for suitability and disclosure. Operations training covers the back-office functions, reconciliation, processing, monitoring, where AI tools are already automating volume work and where staff need to understand the failure modes of the systems they now supervise rather than operate directly.
Why does training compliance and wealth advisory staff for AI differ from generic AI literacy?
Compliance and wealth advisory are functions where an AI-assisted error carries direct regulatory or client-suitability consequences, not just a productivity loss. Training for these roles needs to cover judgment about when not to rely on an AI output, not just how to prompt one, and it needs to be defensible to a regulator after the fact.
A generic AI-literacy course teaches people to write better prompts and to spot obvious hallucinations. That is necessary but not sufficient for compliance officers who must sign off on AI-assisted monitoring alerts, or wealth managers whose AI-drafted portfolio commentary still has to meet suitability and disclosure rules that predate any AI tool. The distinguishing skill in these roles is knowing the boundary of what an AI system can be trusted to do unsupervised, and documenting that judgment in a way that holds up under audit.
This is where compliance training built for audit evidence differs from ordinary skills training: the record of who was trained on what, and when, has to survive a regulator’s request years later, not just a manager’s dashboard check next quarter.
Separate Tool Skills From Oversight Skills
Split every AI course for a regulated function into two graded components: operating the tool, and deciding when to override or escalate it. Most vendors only build and certify the first one.
What does “job redesign” mean in practice for a training team?
Job redesign means rewriting competency frameworks and role descriptions to reflect what a job actually requires once AI tools are embedded in it, not simply adding an AI module to the existing curriculum. It changes what a training team measures, not just what it teaches.
A course catalogue answers “what training exists.” A competency framework answers “what does this role require to do the job safely and well,” and that second question changes once AI takes over parts of a task. If an operations analyst no longer manually reconciles transactions but instead reviews an AI system’s reconciliation output, the job’s core competency shifts from execution to review and exception-handling. A training plan that keeps testing the old execution skill is testing the wrong thing.
This is the part of the MAS pledge that most sector announcements skip. Committing budget to run more AI courses is comparatively easy. Rewriting role descriptions, skills taxonomies and hiring criteria to match a redesigned job, across 23 different employers with 23 different HR processes, is a multi-year organisational change project, which is likely why the timeline runs to 2028 rather than 2027.
How do you coordinate AI training across 23 employers with different LMS and TMS systems?
There is no single system of record across 23 institutions, so coordination has to run on a shared framework and shared reporting standard rather than a shared platform. IBF’s existing role certifying training programmes gives the sector a common reference point that individual employers can map their own LMS or TMS completions against.
In practice, this looks like a hub-and-spoke model rather than a single shared system. Each institution keeps running its own learning platform and its own course content, but all of them report completions against a common set of IBF-recognised programmes and a common skills taxonomy, so MAS and IBF can aggregate progress toward the 80,000 figure without needing every bank, insurer and asset manager to standardise on the same software.
This is the same problem any multi-employer or multi-site regulated programme runs into, and the same one that learning record store architecture is designed to solve: a common data standard for what counts as evidence of completion, independent of which platform generated it, so reporting up to a regulator does not depend on every participant using identical tools.
What does a role-differentiated AI curriculum look like for a regulated employer?
A role-differentiated curriculum assigns different depth and different assessment criteria by function, rather than running one AI course for the whole organisation. Leaders get oversight and accountability content, client-facing staff get judgment and disclosure content, and operations staff get exception-handling content, each mapped to that role’s actual regulatory exposure.
- Tier 1, general AI literacy: every employee, covering what the organisation’s AI tools do, their known limitations, and acceptable use policy.
- Tier 2, function-specific application: role-based modules for wealth advisory, compliance, operations and similar functions, built around real workflows in that role.
- Tier 3, oversight and escalation: team leads and managers, covering when to override an AI output, how to document that decision, and how to coach a team through the same judgment.
This structure mirrors the logic behind a well-built corporate learning system built on role-based pathways rather than a flat catalogue: the platform assigns content by role and tracks completion against that role’s specific requirement, not a single organisation-wide AI course everyone is nominally assigned.
How should you assess AI-oversight competency, not just course completion?
Assessing AI-oversight competency means testing judgment under realistic scenarios, not just confirming a module was watched to the end. A useful assessment presents a plausible but flawed AI output and checks whether the learner catches the flaw, understands why it is wrong, and knows the correct escalation path.
Scenario-based assessment works better here than a multiple-choice knowledge check, because the failure mode being tested is not “does this person know the policy” but “will this person recognise a specific situation as one where the policy applies.” A compliance officer can recite an escalation policy perfectly and still wave through an AI-flagged alert that looks routine but isn’t. Building assessments around real, anonymised near-miss scenarios, not hypothetical ones, closes that gap far more reliably than a knowledge quiz does.
Test The Override, Not The Prompt
Score every AI-oversight assessment on whether the learner correctly declines to act on a flawed AI output, not on how well they used the tool. A learner who never overrides anything has not been tested on the skill that actually matters in a regulated role.
How do you evidence AI training for a financial-sector regulator?
Evidencing AI training for a regulator means keeping a per-employee record that ties a specific completed programme to a specific competency and a specific date, retrievable on demand, not just an aggregate completion percentage. That record has to survive staff turnover, platform migrations and the multi-year gap between training and an eventual audit.
For a pledge running to 2028, the institutions involved need training records that will still be retrievable and defensible well past that date, since regulatory reviews of AI-related incidents can look back years. That means the underlying LMS or TMS needs to capture not just “completed,” but which version of the course, which competency it maps to, and who assessed it, in a format that does not depend on whichever vendor happens to be running the platform at audit time.
Is Singapore’s MAS pledge a template other regulators will copy?
The pledge is likely to be watched closely by other financial regulators because it combines a hard numeric target, a fixed deadline, and named job functions in one government-convened commitment, rather than a vague sector-wide AI literacy goal. Regulators in other financial centres with similarly concentrated banking and insurance sectors are the most plausible early adopters of a similar model.
What makes it exportable is the pairing of a training target with a job redesign requirement, rather than treating the two as separate workstreams. A regulator elsewhere could plausibly borrow the structure, a convened pioneer batch, a shared certifying body, a named deadline, a companion redesign playbook, without needing Singapore’s exact institutions or exact numbers.
What should L&D leaders outside Singapore or finance take from this?
The transferable part of this pledge is not the 80,000 figure, it is the operating model: tie AI training to named job functions and their actual regulatory exposure, redesign competency frameworks alongside the training rather than after it, and build an evidence trail that survives audit and staff turnover. Any L&D leader in a regulated industry can apply that model regardless of sector or geography.
Concretely, that means starting with a skills taxonomy review before commissioning new AI courses, since the courses only make sense once the roles they support have been redefined. It means treating the coordination challenges seen in government and other regulated-sector learning programmes as a preview of what any multi-employer or multi-site AI training effort will face. And it means budgeting for job redesign work, not just course licences, since 2026 corporate training budget benchmarks generally still assume training spend buys content, not organisational redesign.
Conclusion
Singapore’s MAS AI training pledge sets a concrete, dated target, 80,000 employees across 23 financial institutions trained in AI skills by 2028, and pairs it with a job redesign playbook rather than a course catalogue alone. That combination, not the headline number, is what other regulators and other regulated industries should study.
If you lead L&D for a regulated employer, the practical next step is not to wait for your own regulator to issue a similar mandate. Start by auditing whether your current AI training separates tool skills from oversight judgment for your highest-risk roles, and whether your training records could survive a request for evidence three years from now. Both gaps are fixable before anyone asks you to fix them.
FAQ
Q1. What is Singapore's MAS AI Workforce Co-Lab?
The AI Workforce Co-Lab is the name the Institute of Banking and Finance (IBF) gives to the pioneer batch of 23 banks, insurers and asset managers convened by the Monetary Authority of Singapore to train 80,000 local employees in AI skills by 2028. It pairs that training target with a companion Job Redesign Playbook for Financial Services.
Q2. Which 23 banks, insurers and asset managers are part of the pledge?
MAS and IBF have not published a full named list. Reporting on the 24 September 2026 announcement describes the pioneer batch only by category, banks, insurers and asset managers, without naming individual institutions, so this remains unconfirmed at the time of writing.
Q3. How is the 2026 pledge different from the 2025 bank commitment?
The 2025 commitment came from Singapore’s three local banks alone, covering roughly 35,000 employees with no single sector-wide completion date. The 2026 pledge spans 23 institutions across banking, insurance and asset management, covers roughly 80,000 employees, and carries a fixed 2028 deadline plus a Job Redesign Playbook.
Q4. What is the Job Redesign Playbook for Financial Services?
It is a companion document developed by IBF with the Institute for Human Resource Professionals (IHRP) and the Skills and Workforce Development Agency, meant to rewrite competency frameworks and role descriptions to reflect what jobs require once AI tools are embedded in them, rather than simply adding an AI module to existing training.
Q5. Is the MAS AI training pledge mandatory for the 23 institutions?
The source material describes this as a pledge and commitment announced at an IBF sector event, not as binding legislation with stated penalties for missing the target. No enforcement mechanism or penalty for falling short of the 2028 goal is described in reporting on the announcement.
Q6. Which job functions does the training actually cover?
The pledge names three groups explicitly: leaders, wealth managers, and operations staff. Leadership training focuses on oversight and accountability, wealth management training addresses advisers who remain personally liable for suitability and disclosure, and operations training covers exception-handling as AI automates back-office volume work.
Q7. How can a training team outside Singapore or outside finance use this pledge?
The transferable part is the operating model, not the 80,000 figure: tie AI training to named job functions and their actual regulatory exposure, redesign competency frameworks alongside the training rather than after it, and build an evidence trail for that training that survives an audit and staff turnover.
