Responsible AI has moved beyond principles; now we need to make it work at scale

Suzanne Brink

Head of Responsible AI

Suzanne's profile

Alice Genevois

Responsible AI Lead - Consumer

Alice's profile
Alice Genevois

At a glance:

  • Responsible AI needs to be embedded across the entire AI lifecycle, from identifying opportunities and designing solutions through to deployment, monitoring and ongoing improvement.
  • Scaling Responsible AI means moving beyond small teams of specialists, building AI literacy and making responsible decision-making easier for teams across an organisation.
  • Good AI governance can enable innovation rather than slow it down, particularly when controls are proportionate to risk and integrated into existing delivery processes.
  • Organisations need to measure real-world outcomes, not just technical performance, continually testing whether safeguards, human oversight and monitoring are working as intended.

For a long time, conversations about Responsible AI tended to start with principles, policies and governance frameworks. Those foundations matter, but increasingly, the challenge for organisations is how to turn them into something that works in practice and at scale. 

At Lloyds Banking Group, our Responsible AI Centre of Excellence team works with teams across the organisation to do exactly that: turning governance into practical guidance, guardrails, tools and support that can be built into the way AI is designed, developed, deployed and monitored.  

That doesn’t mean responsibility sits with a central team. Business owners remain accountable for the AI they build and use, supported by the framework and specialist expertise of the Centre of Excellence. Some of our Businesses have dedicated Responsible AI Leads who accelerate turning the approach into practice, applying it in the context of real products, customers and decisions. 

Embedding responsible AI in Business delivery matters increasingly as AI scales across an organisation. Responsible AI isn't a tick-box exercise or an approval at the end. It is a delivery discipline that needs to run through the whole AI lifecycle.

The difficult bit isn't writing Responsible AI principles; it's embedding them 

Through our work in Lloyds Banking Group and conversations with organisations outside it, we hear two challenges particularly often: people and scale. 

The first challenge is people 

You can design a Responsible AI framework, but it only works if people understand what it means for them. Organisations need to build AI literacy and bring people with them, particularly where Responsible AI can initially look like additional work rather than something that enables them to make better decisions.  

At Lloyds Banking Group, we have a range of initiatives that address responsible AI literacy, from a mandatory course for all colleagues to joint problem-solving sessions for Responsible AI focal points in Business.  

The second challenge is scale 

Central expertise remains important as AI scales, but responsible AI at scale cannot depend solely on having the right experts in the room. We need to make it easier for everybody else to make responsible choices too.  

A Responsible AI Centre of Excellence turns specialist expertise into practical approaches that others can use: clear guardrails, guidance, tools and reusable patterns. Specialist support, directly from the Centre or Responsible AI Leads in Business, are only brought to cases that genuinely require it.  

Make the responsible choice the easy choice

There are three lessons from our own journey that I find myself sharing often:

1. Start early

Responsible AI shouldn’t be an approval stage at the end of development. By then, many significant decisions may already have been made, whereas bringing responsible thinking into a project early means it can influence the design itself. 

Teams should think about potential harms while the solution is still taking shape, incorporate appropriate guardrails into their plans and identify the evidence they will need to demonstrate that their controls work.

2. Integrate Responsible AI into delivery

People building AI should not need to navigate a completely separate universe of policies and processes to understand what responsible development requires of them. As far as possible, the right questions and controls should appear within the journey teams are already following. That requires Responsible AI expertise, engineering and business teams working together, translating governance into practical controls that are embedded in the end-to-end life cycle.  

3. Encourage reusability

If an organisation finds it is solving the same Responsible AI problem repeatedly, this signals an opportunity to create a repeatable pattern. Repeatable patterns, usable assets and governed components can reduce duplication and give teams a stronger starting point, while still allowing the approach to be tailored to the risks and context of an individual use case. 

That is ultimately the principle behind much of this: make it easier for people to do the right thing. These are some of the practical lessons we explore in our Responsible AI guide, drawing on real experiences of applying Responsible AI within Lloyds Banking Group.

Good governance should help you move faster 

There can still be an assumption that governance and innovation sit at opposite ends of a spectrum. Strengthen one and you inevitably constrain the other. We don't think that is the right way to look at it. 

A colleague once compared governance to the brakes on a car. The brakes may technically stop the car, but that isn't their only effect. Having good brakes actually gives you the confidence to drive faster. Responsible AI can work in much the same way.  

If a team knows what it needs to do, understands where the boundaries are and has reusable, governed components available, there is less uncertainty and less need to solve the problem from scratch. This can give teams the confidence to move forward with controls in place. 

The other important part of this is proportionality. Two AI systems with very different purposes and risk profiles should not automatically be expected to follow identical requirements. The consideration given to a use case and the controls applied should reflect its materiality and risks. 

Responsible AI therefore isn't about creating friction everywhere. It is about applying the right challenge in the right place.  

Industry and policymakers have different perspectives, and we need both 

We work with organisations and policymakers outside of Lloyds Banking Group, with our leaders for example being member of the Government’s Digital Service Responsible AI Advisory Panel and Bank of England’s AI Consortium. This allows us to contribute our lessons learned and also challenge our own thinking. Those implementing AI at scale bring evidence about what happens when principles meet operational reality: where the difficult decisions arise, what can be measured, what has worked and where approaches can fail. 

This exchange is also valuable in the other direction. Policy discussions can introduce a different set of starting questions, including societal impact, accessibility, legitimacy, public trust and data sovereignty. They don't replace the questions organisations already ask, but they can broaden the lens through which we look at an AI system. 

Where those perspectives meet is around evidence: What decisions did you make? What potential harms did you consider? Why did you choose particular guardrails? What evidence tells you that they are working? 

These are important questions regardless of whether AI is being deployed by a bank, a business or a public body.  

 

Artificial Intelligence at Lloyds Banking Group

We’re reimagining how we operate by harnessing the full potential of AI – embedding it across our business to drive smarter decisions, faster outcomes, and better experiences.

Visit the AI hub

We all need to get better at measuring outcomes, not activity 

Ultimately, assessing the success of an AI system and monitoring its performance has to take us back to the outcome the AI was designed to achieve for the people using it. Technical performance matters, but it doesn't tell you on its own whether a customer is getting the outcome you intended. 

That means starting with the user or customer: what does a good outcome actually look like for this particular use case, and how will we know whether we're achieving it? There won't always be a single metric that answers that question. 

Imagine an AI-enabled health application. If somebody logs in every day, is that evidence the application is succeeding? Perhaps. But perhaps frequent use isn't the outcome you ultimately want. If somebody stops using it, has the application failed? Or did it help them achieve what they needed to achieve? 

That’s why Responsible AI also can't finish at deployment. The assumptions made during design need to be tested against what happens in the real world, and organisations need to be prepared to adapt their controls and measures as they learn.  

The aim is for Responsible AI to become... boring! 

So what would good Responsible AI look like in three years? We hope it feels much less exceptional. 

As AI evolves, dedicated responsible AI expertise will remain essential in identifying emerging risks, developing the Group approach and helping ensure governance and controls remain fit for purpose. 

But that expertise should also become increasingly easy to put into practice. Clear guardrails, established patterns, reusable approaches and capability within individual businesses can make good responsible practice part of the way teams simply do AI. 

That is the model we are continuing to build towards: central expertise that evolves as technology and risks change, alongside business teams equipped to apply it, learn from real-world outcomes and feed those lessons back. Our new Responsible AI guide shares practical lessons from Lloyds Banking Group's experience, including real examples of how teams have turned responsible principles into decisions, guardrails and controls.

 

Responsible AI at a glance

  • Responsible AI brings together governance, risk management, human oversight, technical controls, monitoring and assurance to help organisations adopt AI with confidence, while retaining the trust of customers, colleagues and wider society. 

    Take a look at our Responsbile AI Guide to find out more.

  • AI can introduce or amplify risks that conventional technology controls may not fully address. Responsible AI helps organisations understand those risks, establish appropriate safeguards and evidence that AI systems continue to work as intended.

    Take a look at our Responsbile AI Guide to find out more.

Related content

How is Lloyds Banking Group performing in 2026?

William Chalmers and Jas Singh OBE

As we publish our 2026 half year results and the next phase of our strategy, we continue to support customers and businesses, simplify how we work, and invest in the technology and skills that will define the future of financial services.

Read their article