As artificial intelligence increasingly cuts across and blurs what have traditionally been quite clear, functional boundaries, investment organisations are grappling with the question of how to maintain optimum oversight and accountability.
For decades, investment organisations have been structured around relatively clear lines of accountability. The chief investment officer owns investment decisions; the chief operating officer owns operations; data teams own data; technology owns systems; and risk provides oversight.
That traditional model was developed for good reason. Functional accountability makes organisations manageable. But AI is a capability, rather than a conventional function. It cuts across investment, operations, technology, data, risk and, increasingly, every corporate function.
Consider the example of an AI model that is implemented by a technology or operations function, using data governed by another function, that subsequently informs an investment decision. The technology sits in one function, the data in another and the decision in a third.
This is not a hypothetical scenario. At the recent 11th Investment Data and Technology Summit in Sydney, an audience poll found that the use of AI agents in investment workflows is rising rapidly – 29 per cent of attendees already had AI agents in production, with a further 60 per cent having agents in pilot or proof of concept stage.
Regulators are concerned
In April 2026, the Australian Prudential Regulation Authority (APRA) warned that governance, risk management and assurance practices across banks, insurers and superannuation trustees are not keeping pace with the pace at which AI is being adopted.
A targeted engagement involving a select group of large banks, insurers and superannuation trustees that was conducted by APRA in late 2025 highlighted a tendency among entities to treat AI as “just another technology”. That puts organisations at risk of missing key differences, such as the “distinct characteristics of predictive systems, adaptive behaviour in models, ethical considerations such as inherent bias, and privacy and data risks”, the regulator said.
APRA said that across the AI lifecycle, it has identified gaps including weak controls over post deployment monitoring, weak model behaviour monitoring, change management, and decommissioning of AI capabilities. The regulator also noted that traditional change management and assurance, while in place, are not sufficient for dynamic AI solutions.
According to Joel Grant, founder and principal at Jolly Rambler, an investment technology governance consultancy firm, AI is accelerating a governance challenge that has been increasing the more investment portfolios have become dependent on technology and data.
“AI is change on steroids, compressing timelines and collapsing the distance between the portfolio and the technology program that serves it,” Grant wrote in a recent paper (https://www.jollyrambler.com.au/own-it-like-a-cio.html).
He agreed that organisations are struggling to adapt their operating models at the same pace as technology.
“The profound effects are not yet fully visible in how large asset owners are structured or how roles are defined. Organisations adapt slowly. In that gap – between rapid advances in AI capability for investment programs and how institutions are responding – governance vacuums form,” Grant said.
An audience poll at the event found that the most significant shifts in operating models were in ‘how work gets done’ – not in redefining accountability and governance. Specifically, 74 per cent of respondents ticked increasing automation and AI-enabled workflows against 18 per cent that flagged changes in redefining ownership, oversight and accountability as AI becomes embedded in those workflows and decisions. Eight per cent said skills, roles and organisational boundaries had changed most significantly.
Indeed, across many firms, accountability lines remain as they ever were. For example, the CIO remains responsible for investment decisions, regardless of whether AI models are used to inform those decisions.
At IFM Investors, a global data, digital and AI function has been established within the operations, data & technology division that reports into Amy Diab, the firm’s chief operating officer. The function’s purpose is to assess value, prioritise and enable the adoption of opportunities across the enterprise, in accordance with the enterprise AI framework. While the AI function is responsible for designing the AI operating model and governance framework, including essential capabilities to activate AI at scale, Diab said it cannot be the owner of every AI-related decision.
“Business executives remain accountable for the decisions made and the outcomes within their areas, including where they rely on AI,” Diab said.
“If an AI model provides an input into an investment decision, the relevant investment
decision-maker remains accountable for that decision. The fact that AI was used does not turn it into a technology, data or risk decision, nor does it transfer accountability to an AI function,” she said.
That is also the view taken by Grant, who said that in turn, a CIO’s role has expanded quite significantly.
“The CIO of 2026 is the CEO of the investment program — responsible for investment outcomes, but also for the business of running it: leading a significant team, organisational leadership, culture, and the capability inputs that serve the portfolio, including technology and data,” he said.
Grant said that while portfolios have always required data, the degree to which portfolios are now mediated by and reliant on technology, data and data infrastructure, means that the technology program that produces the data and infrastructure to inform decisions must be a focus for CIOs and senior members of the investment team.
“The importance of investment teams owning these investment technology and data decisions has never been greater,” Grant said. “The CIO is already accountable for the outcomes that depend on the technology program, and they now need to own the program that produces those outcomes.”
A human in the loop is not enough
Conversations about accountability typically refer back to the importance of having a ‘human in the loop’ who is ultimately responsible for decisions. Indeed, the phrase itself was mentioned 26 times in various discussions at the recent event.
There was, however, an important qualification. A human in the loop is only meaningful if that person has access to the relevant inputs, understands the limitations of the system, can challenge its conclusions and has sufficient authority to intervene.
In a recent conversation with Fund Business, Wendy Antaw, chief technology officer at Perpetual, suggested the situation is more nuanced than simply appointing a senior executive with blanket accountability for everything that is decided.
“You have to be quite deliberate about that. It’s when you’re not [deliberate] that issues typically come up, or mistakes are made,” she said.
Organisations need to consider the particular AI solution, its inherent weaknesses and its risk profile, and then determine who is best placed to manage those risks, she said.
“If it’s a user-developed AI solution, then it’s very clear that the user is accountable, but in some cases, there might be multiple accountabilities, involving the data owner or the technology department,” she said, adding that there is no universal one-size-fits-all answer.
“Organisations need to consider who’s in the best position with the skills and proximity to do that and make it explicit and consider that on a case-by-case basis,” she said.
Humans in the loop are also only effective so long as they actually exercise accountability.
As one speaker at the event quipped: “There can be almost an assumption that having a human in the loop means that everything is going to be fine. But if that person is just there clicking approval buttons that is not making your processes verified and safe. If it’s not being checked and challenged, you don’t actually get the right outcome.”
Accountability requires expertise
Accepting the importance of human oversight over increasingly automated processes raises a longer-term question about how organisations develop and retain the expertise required for such effective oversight.
In a traditional model, junior members of staff are hired to do the groundwork, thereby building skills and judgement that ultimately turns them into experts.
AI removes much of that underlying work which – on a short-term basis – would seem to be an entirely rational productivity trade-off. Organisations do not need people to perform manually what machines can do more efficiently.
The problem with that approach is that this also removes part of the learning process through which investment professionals develop the ability to recognise a bad answer. In the worst-case scenario, it is those people who end up responsible for challenging increasingly sophisticated AI systems.
Discussions at the conference highlighted the importance of designing AI-enabled workflows in such a way that learning pathways are not eliminated.
“We should be using AI to accelerate learning, not to eliminate it,” one panelist said. For early-career investment professionals in particular, organisations may need to reconsider how judgement is developed when machines perform much of the work through which previous generations acquired it.
That may be the more fundamental governance challenge posed by AI: not determining who is accountable, but ensuring that accountability remains meaningful as the processes behind investment decisions become increasingly distributed, automated and opaque.
