Data center corridor filled with servers illuminated by bright white light

Featured image by Jenna O’Malley/PitchBook News

Capital Pool

Allocators add AI to ESG concerns

“The biggest asset owners, the biggest sovereigns, will have exposure to the knock-on effects.”

LPs are starting to add a new section to their due diligence: AI ethics.

ESG might’ve fallen out of the limelight, but as AI regulation moves through statehouses in California and New York, large pension systems, sovereign wealth funds and corporates are taking the risks of AI increasingly seriously.

Asset managers around the world are discussing how to ensure that they, both directly and through their fund stakes, are invested in ethical AI systems.

“It’s a whole new set of considerations,” said Natasha Allen, partner at Foley’s Silicon Valley practice.

For some, it’s already set in stone in their due diligence. Canadian insurance giant Manulife shared its responsible AI principles publicly earlier this year. StepStone Group rolled out responsible AI in its due diligence in 2024. And Norway’s sovereign wealth fund is well known for its AI accountability measures.

“This is actually a very active conversation with our LPs,” said Lila Preston, who leads the growth equity strategy at Al Gore’s sustainable investment firm, Generation Investment Management.

This article originally appeared in the Capital Pool newsletter.
Subscribe to the newsletter here

Behind the scenes, many more LPs are figuring out how to ensure they’re investing in AI systems that are safe, don’t reinforce biases or violate copyright, and that—crucially—don’t pose a fundamental risk to the livelihoods of their investor base.

Public pension funds have a particularly complex task: ensuring that their AI bets don’t erase the jobs—everything from clerks to assistants to teachers—that comprise the funding resource for their pension.

“You’re going to essentially eradicate your own investment base if you invest in things that may at some point in time replace their jobs,” Allen said.

Pension funds are used to ESG. But ensuring AI bets don’t hurt their own plan’s participants is an entirely new concern.

“It’s much more nascent, it’s only in the last couple of years that these pension funds have started looking at the jobs consequences,” said Mike Kubzansky, CEO of philanthropic investment firm Omidyar Network.

The firm, which has collaborated with several large pension funds has a list of “essential questions for responsible AI diligence.” The questions are span algorithmic accountability, contingency measures such as kill switches, and monitoring hallucinations.

AI can help build positive systems: Founders are applying it to everything from making supply chains more efficient to optimizing data workflows. But simultaneously, it poses huge risks: in rising electricity and water consumption, reinforcing discrimination, data privacy and exposing systems to novel cybersecurity threats.

Sovereign wealth funds tread a delicate line—many have specified job creation mandates in their charters. Ireland’s sovereign development fund has a mandate to support employment in Ireland, while Kuwait’s Ciyada Development Fund was launched in 2023 to promote the local economy.

Universal owners are especially sensitive to AI risks, because those externalities will almost inevitably show up somewhere in their vast portfolios.

“The biggest asset owners, the biggest sovereigns, will have exposure to the knock-on effects,” said Preston, adding, “if there’s social unrest, the country will feel it and it will come through to the pensioners of that country itself.”

Conducting due diligence on something so new and evolving poses a fresh challenge. Most early-stage companies don’t have in-house counsel, let alone a board of directors.

There are still early signals that investors can reference, however, like having people on staff tasked with AI safety.

Kubansky’s Omidyar Network, alongside the Ford Foundation and Nathan Cummings Foundation, invested in large language model creator Anthropic last year and were reassured by the company’s commitment to an AI safety charter.

And there are some red flags that LPs can be aware of, too: For one, “anyone who says that their business is not impacted by AI,” according to Suzanne Tavill, partner and head of responsible investing at StepStone Group. Or, “anyone who devolves responsibility of ‘managing’ AI to an external consultant,” she added.

As more advanced AI agents come online and deep-pocketed startups pursue the quest for AGI, or human-level intelligence, allocators will continue to encounter novel philosophical questions and the need for guardrails.

One field of inquiry that LPs will rapidly have to pay more attention to is that of agentic control and management. Mechanisms like “human in the loop,” user overrides and user consent guardrails can help safeguard against those risks.

“We’re looking for the agent models that have a duty of loyalty to [the user], not to the agent software-maker,” said Kubzansky.

Featured image by Jenna O’Malley/PitchBook News

Learn about our editorial standards.