‘AI-washing’ risk grows as tech due diligence on managers lags

Asset owners have been warned to guard against “AI-washing” among managers as the pace of development for artificial intelligence models has significantly outpaced the refresh-rate of most due diligence frameworks.

According to a recent report from the Institute of Sovereign Investors, an international network of academics and investors, allocators today are pitched an unprecedented number of technical, AI-driven investment strategies from managers. However, not all allocators have specialist staff that can spot the AI-related red flags. A real challenge is distinguishing proprietary models from those that are “a superficial wrapper around a commercial API”.

Paul O’Brien, former ADIA deputy chief investment officer and member of the institute’s international advisory council (IAC), and a co-author of the report, said few managers can have truly unique, firm-specific data assets in “a world of ubiquitous AI”.

“This is especially the case in public markets, but even in private markets how do you keep proprietary information confidential?” he told Top1000funds.com.

The report has identified public equity as the asset class most prone to AI disruption, with large language models (LLMs) already extracting signals from transcripts, filings and macro commentary at scale.

Liquid fixed income and systemic hedge fund strategies are also at fair risk, with AI already excelling at credit screening and macro processing and offering a direct upgrade to “quant pipelines” – the end-to-end quant investment processes – the report said.

Private equity is comparatively the safest from AI disruption, but the risk level is still “moderate”. AI is already cleaning and structuring fragmented private data and enhancing due diligence, though value creation is still relationship-driven.

Kristian Flyvholm, the institute’s chief executive and former CEO of the IFSWF, said that to avoid the risk of managers “AI-washing”, allocators need to check if the governance structure, leadership commitment, technology architecture and talent composition at a manager have the institutional capacity to support a true moat.

This means asking questions about whether a manager has a resilient AI strategy in place, who will be responsible for any failure of AI systems and what the backup plans are during system outages, alongside understanding the technical aspects. Also, investment management agreements needs to be clear on potential liabilities related to AI-related failures.

“The due diligence check list in fund selection is clearly going to change, and we are quite clear that, in our view, in a few more years, there’s going to be a market standard of AI due diligence that everybody incorporates into their investment and external manager framework,” he said.

Manager selection critical

The prevalence of AI integration also means it’s even more important to pick top-quartile managers, as O’Brien argued that many traditional forms of alpha that are gained from information asymmetry will be “eliminated”.

“Every manager can have AI agents acting like a large team of PhD-level quants and finance researchers. But this leads to the really fundamental question: how does AI affect market-level price discovery and efficiency?” he said.

“In my opinion AI will make access to information and analytic power much cheaper and widespread, thus eliminating many traditional forms of alpha. Successful active management becomes tougher, and allocators can rely more on cheaper passive strategies.”

It may seem that larger managers have the natural advantage in the AI race, having bigger budgets to develop proprietary AI models and databases, but Flyvholm said boutique managers can also reap a lot of “operational alpha” from AI-driven efficiency gains. These relate to more efficient processes and more data to augment portfolio managers and decision-making.

A recent Stanford paper highlighted that budget and governance constraints have long pushed allocators to offload bold innovation to external managers, who are commercially incentivised to pursue them, which the paper’s author, Ashby Monk, argued is instilling short-termism in pension management.

But O’Brien highlighted that AI has the potential to shrink the resource gap between public-sector asset owners and asset managers.

“Allocators don’t have to do ‘tech innovations’. They know what good investment results are. They only need to tap into the burgeoning supply of new AI-enabled tools,” he said.

“Allocators also can work together. They are not competing with each other. They are competing with the market for good returns, and with managers for lower costs and better transparency.”

For example, asset owners can explore assigning AI agents to track the performance of their managers and even monitor managers they did not hire. They can also expand the scope of scale of due diligence and screen more candidate managers, O’Brien said.

But ultimately, how much benefit an organisation gets from AI will depend on their risk appetite, said Reza Mahmud, who is also on the Institute’s International Advisory Council and co-author of the report.

“We’ve seen AI move from being a siloed experiment in an organisation, into the bolder, enterprise level transformation. But not everybody has that risk appetite or the capability to do AI at scale. By the end of the day, AI is a way of augmenting humans to work smarter, while the AI agents work harder,” he said.