On 22 June 2026 Banco Santander extended AI tools to every one of its 185,000 employees worldwide. Until then about 40,000 staff had access. Everyone now has Microsoft Copilot for everyday work, alongside tools from OpenAI, Anthropic and Google, for preparing analysis, finding information, summarising documents and improving customer conversations. AI training is mandatory for all employees from 2026, with a core module on responsible AI, a commitment the bank first published in August 2025. Santander reports €35 million of measured value from AI in the first quarter of 2026, expects more than €200 million for the full year, and has set a target of more than €1 billion in combined revenue gains and cost savings between 2026 and 2028. Around 17,000 of its developers use agentic AI tools and 40 percent of the bank’s code was written by AI in June. AI now supports more than 40 percent of contact centre interactions. The stated aim is to be an AI-native bank by 2027.
Santander is not alone. Citi began mandatory prompt-writing training for its 175,000 employees in October 2025. JPMorgan has made AI training a compulsory part of onboarding for every new employee since 2024. Bank of America reported in April 2025 that more than 90 percent of its 213,000 staff use AI tools in daily work.
Those are the facts. What follows is about what an employer’s AI training can cover, what it cannot, and who is responsible for the rest.
What the bank is teaching
Santander has said what its mandatory training will do. It will be delivered across roles and markets so that every employee can use the tools safely and effectively, and it will carry a core module on responsible AI. That is the right shape for an employer. It teaches people how the tools work, what they must not put into them, how the bank’s policies apply, and where the limits are. It also discharges an obligation the bank has in law. Article 4 of the EU AI Act asks any organisation deploying AI to see that its staff have a sufficient level of AI literacy, and a mandatory training programme for all staff with a responsible-use module is a reasonable answer to it.
Call this the employer’s layer. It is owed by the bank to its staff, its customers and its regulators. It is about the tool and the workplace. Every large employer will have one within a year or two, whether it is called AI literacy, responsible AI, or something else.
The layer the bank cannot teach
Inside 185,000 employees are thousands of people who carry a second set of obligations that do not come from Santander. Chartered accountants in the finance function. Actuaries in the insurance arm. Solicitors in legal. Compliance officers, internal auditors, financial advisers, company secretaries. Each holds a professional qualification, belongs to a professional body, and is bound by that body’s code. The code binds them personally, follows them from employer to employer, and can end their career if they breach it.
Those codes were written before the tools existed, and they apply to them anyway. Four duties in particular.
Confidentiality. A chartered accountant’s duty not to disclose client or employer information is not satisfied by the bank’s data policy. It is the accountant’s own duty, and if information ends up somewhere it should not, the professional body asks the accountant, not the bank.
Verification. A professional who relies on an AI output in their work is responsible for that output as if they had produced it themselves. Most accountancy and legal codes now say this in some form. A summary that misses a figure or a draft that cites a source that does not exist is the professional’s error.
Disclosure. Whether, when and to whom a professional must say that AI was used is a question of professional ethics, not of the bank’s tool policy. The answer differs between a board paper, a client letter and a regulatory filing.
Judgement. The duty to bring an inquiring mind and professional scepticism to any information, including information produced by a machine, is the core of what a professional qualification certifies. It cannot be delegated to a tool and it cannot be delegated to an employer’s training either.
An employer’s module is not designed to teach these, and no responsible employer would claim to certify them. It is not the bank’s place to say what a chartered accountant’s duty of confidentiality requires. That belongs to the body that granted the qualification.
Two layers, two owners
So there are two layers of AI training for any regulated professional in a large organisation, and they have different owners.
The employer teaches the tool. The profession teaches the duty.
The employer teaches how the tool works, what the organisation permits, where the risks sit, how to use it well. The profession teaches what a member is personally accountable for when a tool is in the room, and what they must do before, during and after using one.
The first layer is being built at scale now. Santander, Citi, JPMorgan and Bank of America are four examples from one sector in one year. The second layer is, in most professions, still to be built. Many bodies have issued guidance. The next step, a CPD programme that puts a member in a realistic situation with an AI tool in it and asks them to decide what their duty requires, is where the work now is.
That gap will not stay open long, because the employers will close it from their side if the professions do not. A bank that has trained 185,000 people on the tool will, sooner or later, ask its chartered accountants to show that they have been trained on the duty. If the professional body has that programme, the member completes it and the body’s certificate is the evidence. If the body does not, the bank will write its own, and the working definition of a professional duty will have been written by an employer.
The bodies that move first will find the employers grateful. The employer’s layer is expensive to build and hard to keep current. The duty layer is the one part of the problem the employer cannot solve, and a body that hands its members a credible answer to it is giving every employer of those members something they need.
What the duty layer looks like
It is not a webinar on what large language models are. Members have had those. It is a situation.
A member of a finance team is asked for a board paper by Thursday. The tool produces a good draft in minutes. Somewhere in it is a figure that does not reconcile, a claim about a competitor that cannot be sourced, and a paragraph that reads well and is subtly wrong. The member has to decide at several points, and every wrong choice is the natural thing to do under time pressure. Send it because it looks finished. Check the numbers and not the claims. Present it with a caveat. Tell nobody that a tool drafted it.
The right answers come from the code, not from the bank’s policy. The programme’s job is to put the member in the situation before it happens for real, let them decide, show them the consequence, and record what they decided. Done well, it produces something the employer’s layer never can: evidence, under the professional body’s own name, that a named member exercised professional judgement with an AI tool in the room.
The duty layer, ready to license
AI for Professional Practice is that duty layer. It is a CPD programme built for professional bodies to put in front of their members, under the body’s own name. Module 1 is live now: a twenty-minute scenario on AI and professional duty in which the member makes real decisions about confidentiality, verification, disclosure and judgement, sees where each decision leads, and receives a scored result and, at the pass mark, a certificate. Any education team can complete it today at learnframe.com/course/ai-for-professional-practice/module-1/, and the programme is available to license at learnframe.com/license-ai-for-professional-practice.
Santander has built the first layer for 185,000 people. The bodies whose members work there own the second. It can be in place before the bank asks for it.
Want to know what your own programme could tell you that it currently does not? The Programme Design Diagnostic is a fixed-fee, independent read on a single programme: a board-ready diagnosis and a costed build scope, in two to three weeks.
Sources: Banco Santander, “Santander’s data and AI-first strategy accelerates through OpenAI collaboration”, August 2025, including the commitment to mandatory AI training from 2026 with a responsible AI module. Finextra, 24 June 2026, Santander extends AI access to all 185,000 employees. PYMNTS, 22 June 2026, on the €1 billion target for 2026 to 2028. Crowdfund Insider, 22 June 2026, on the €35 million first-quarter figure, the 17,000 developers and the 40 percent of code. Computer Weekly, 13 August 2025, Santander will make AI training mandatory for all staff in 2026. Fortune, 1 October 2025, Citi begins retraining 175,000 employees, including the JPMorgan and Bank of America figures. All retrieved 8 September 2026.
For what Article 4 requires of employers, see EU AI Act Article 4: AI literacy applies from 2 February 2025. For what AI literacy means in plain terms, see AI literacy in plain English. For another large employer’s AI training at scale, see Accenture’s AI upskilling programme. See more insights from LearnFrame.