For the first time, the Belgian Ombudsman for pensions has reported complaints involving artificial intelligence and warned citizens not to rely on AI chatbots for complex pension information.

This reveals a broader reality: pension providers are no longer the first source of information for many members. AI has become a new intermediary between pension providers and their members

AI is changing how people access pension information

Pension institutions are subject to extensive information requirements and have invested heavily in member communication. Savers receive annual statements, visit scheme websites, call helplines, attend information sessions or seek regulated financial advice. Historically, pension funds communicated directly with participants.

Today, information increasingly passes through an algorithm before reaching the individual, as many members are asking ChatGPT, Gemini, Copilot, or other AI tools for answers instead of contacting their pension provider.

Pensions are certainly not the only area where this is happening. Increasingly, people are using AI to interpret medical symptoms, legal documents, financial products and tax issues before consulting a professional.

This raises an important question: Is the European pensions industry prepared for a world in which retirement decisions are increasingly influenced by artificial intelligence?

New risks for pension savers

The Belgian experience suggests that caution is warranted. According to the Ombudsman, pension institutions were, in most cases, providing the correct information, while AI-generated responses had introduced confusion or inaccuracies.

This reflects a fundamental challenge.

General-purpose AI systems are designed to generate plausible responses based on patterns in data. They are not pension specialists. They may misunderstand scheme rules, confuse national pension systems, rely on outdated information or present uncertain conclusions with unwarranted confidence.

The challenge is not merely that AI may occasionally provide incorrect information. It is that many members increasingly trust AI as their first source of information. Pension funds, therefore, face competition not only from misinformation but also from a new digital intermediary between themselves and their members.

The issue is particularly relevant in Europe, where pension systems differ significantly between member states. An AI system trained primarily on information from one country may provide answers that are entirely inappropriate in another.

Occupational pensions, personal pensions and statutory pensions operate under different legal, tax and social frameworks, making accuracy especially important. A member asking a seemingly simple question about retirement age, pension taxation or transfer rights may receive an answer that is perfectly correct in one country but entirely wrong in another.

The risks are especially significant because pension decisions are often complex, infrequent and, in some cases, irreversible. A misunderstanding about retirement age, transfer rights, survivor benefits, contribution levels or decumulation options can have meaningful financial consequences for individuals.

At the same time, pension providers remain accountable for their own decisions and communications, irrespective of information that members may have obtained elsewhere.

Designing pension information for humans and AI

Artificial intelligence is here to stay and the solution is not to discourage its use. Members will continue to use AI because it is convenient, available around the clock and capable of providing immediate responses. The real question is how pension providers should respond.

The first responsibility of the industry is to ensure that trustworthy pension information is easier to access than inaccurate information. Pension funds cannot control what external AI systems say, but they can improve the quality, accessibility and clarity of their own communications.

Information that is clear, consistent and written in plain language is more likely to be correctly interpreted both by members and by AI systems.

This is where organisations such as Quietroom in the United Kingdom have demonstrated valuable leadership. Their work has highlighted that pension communications must increasingly be designed not only for human readers but also for AI-driven search and information tools.

Their research has shown that when pension information is difficult to locate, fragmented or overly technical, AI systems are more likely to generate inaccurate answers. Conversely, well-structured and accessible information improves the quality of AI-generated responses.


Learning from emerging industry practices

Several pension organisations are already taking practical steps. Ford's UK pension schemes have developed PensionsChat, an AI assistant trained specifically on scheme documentation.

Rather than leaving members dependent on generic AI tools, the scheme offers a controlled environment where responses are grounded in official scheme information.

This recognises an important reality: If members are going to use AI anyway, pension providers should consider offering safer and more reliable AI tools themselves.

Several Nordic pension systems have long provided consolidated digital pension information through services such as MinPension in Sweden and PensionsInfo in Denmark.

Although these platforms were created before generative AI, they provide an important model: a single trusted source of pension information. The next generation of these types of services may act as trusted AI-powered assistants that help individuals understand their retirement position across multiple pension arrangements while ensuring that answers are based on verified pension data rather than internet searches.

The future may lie not in generic AI but in AI embedded within trusted pension information ecosystems.

Other pension organisations are using AI to improve member engagement and communication. Dutch pension administrator APG, for example, has used AI and predictive analytics to better understand participant needs and deliver more relevant communications.

Such initiatives are not about replacing human interaction but about helping members find the right information at the right time through a combination of AI-enabled tools and human support.

AI governance and fiduciary responsibility

From a pension fund perspective, the issue ultimately relates to fiduciary duty. Pension institutions have a responsibility to act in the best interests of their members and beneficiaries.

While they cannot control external AI systems, they can ensure that their own information is accurate, accessible and understandable.

Artificial intelligence can help answer routine questions, guide members through administrative processes and improve accessibility. However, important decisions involving retirement timing, benefit options, transfers, decumulation choices or complex personal circumstances should continue to involve human expertise and judgement.

Pension providers should ensure that members can easily escalate from AI-generated information to qualified support whenever necessary.


Preparing for the AI Act's AI literacy requirements

For pension funds, AI literacy should not be viewed merely as a compliance exercise. Understanding how AI systems generate responses, where errors may occur and when human intervention is required is increasingly becoming an element of good governance and risk management.

Staff responsible for member communications should understand the risks of AI-generated misinformation. Investment and operational teams should understand the limitations of AI-driven analytical tools. Trustees and board members should be sufficiently informed to oversee the deployment of AI within the institution and to assess associated risks and controls.

The objective is not to turn pension professionals into AI specialists. Rather, it is to ensure that those using AI systems possess an appropriate level of understanding to use them responsibly, identify potential shortcomings and maintain effective human oversight.

In this respect, AI literacy is closely aligned with the broader fiduciary responsibility of pension institutions to act in the best interests of their members and beneficiaries.

In addition to the practical challenges associated with AI-generated pension information, pension institutions should also be aware of their obligations under the European Union's Artificial Intelligence Act.

Article 4 of the AI Act, which entered into application on 2 February 2025, requires providers and deployers of AI systems to ensure a sufficient level of AI literacy among the persons dealing with AI systems on their behalf. While the precise application of this requirement depends on the circumstances, pension funds that deploy AI systems in areas such as member communications, administration, customer service, investment operations or internal decision-support tools may fall within its scope.

The supervision and enforcement framework applicable to these obligations will become operational from 3 August 2026 onwards. Pension institutions should therefore already be considering how to ensure that staff, management bodies and relevant service providers have an appropriate understanding of the capabilities, limitations and risks associated with the AI systems they use.

Importantly, the European institutions have recognised the need for a proportionate approach. The forthcoming Digital Omnibus package is expected to clarify and moderate the expectations associated with Article 4. The revised text places greater emphasis on taking reasonable measures to promote AI literacy while recognising the differing roles, resources and circumstances of organisations deploying AI systems.


The Future: Combining AI with human expertise

As AI becomes increasingly embedded in pension administration, communications and investment processes, the industry's challenge will be not only to harness the benefits of AI but also to ensure that its use remains transparent, accountable and subject to appropriate human judgement.

The industry's responsibility is therefore not to intervene in every saver decision or to become a gatekeeper of AI usage. Rather, it is to create an environment in which members can make informed decisions based on accurate information, supported by appropriate governance, transparency and human oversight.

The future of pension communication will be a combination of both AI and human expertise. The pension institutions that succeed will be those that use AI to improve accessibility and understanding while preserving human judgement, accountability and trust.

Building trust in the age of AI will require not only accurate information and human oversight, but also a sufficient level of AI literacy throughout pension institutions, as increasingly recognised by the EU regulatory framework.

---