TOKYO -- Pension funds and asset managers worldwide are stepping up scrutiny of how portfolio companies manage artificial intelligence, aiming to prevent their value from deteriorating by encouraging proxy voting against the election of directors and other methods.

The California Public Employees' Retirement System (CalPERS), one of the largest public pension funds in the U.S., included AI-related provisions for the first time in its proxy voting guidelines revised in April.

They state that CalPERS may withhold support for director election proposals at shareholder meetings if board oversight of AI-related risks is inadequate or has failed to function effectively.

The fund will decide whether to apply those provisions on a case-by-case basis. A CalPERS representative said it generally looks at whether a company has suffered financial losses or reputational damage as a result of failing to appropriately oversee AI-related risks.

Pensions UK, a British industry association, said in its revised 2026 voting guidelines that investors should consider voting against director candidates and other proposals if adequate governance of AI risk is not in place.

"The 2024-25 voting season has seen a marked increase in shareholder resolutions related to AI and cybersecurity governance, underscoring the urgency for investors to develop clear expectations around board oversight, responsible use frameworks and risk mitigation," the guidelines state.

The New York State Common Retirement Fund added AI as a stand-alone category in its proxy voting guidelines revised in 2026. The new guidelines place particular emphasis on board accountability and measures to manage AI hallucinations, and indicate that the fund will support shareholder proposals calling for greater transparency in AI-related disclosures. HSBC Asset Management also added AI-related provisions to its proxy voting guidelines for the first time this year.

Institutional investors have traditionally exercised voting rights based on factors like corporate performance as well as environmental, social and governance initiatives, but movements to add AI-specific criteria are spreading in Europe and the U.S.

At Apple's 2024 annual shareholders meeting, a shareholder proposal called on the company to disclose its use of AI and its ethical guidelines governing the technology. Apple opposed the idea, citing reasons including the protection of confidential information, and it was voted down. But support for such proposals may rise going forward.

In Japan, while major asset managers consider whether corporate value is damaged by scandals or other incidents, none so far have incorporated AI as a stand-alone category in their proxy voting guidelines.

"Rather than treating AI as a separate issue, we view it more broadly as part of a company's risk management framework and look at whether portfolio companies have properly established systems for information management and cybersecurity," said Masanobu Endo of Asset Management One.

AI has rapidly become entrenched across all parts of society. The number of generative AI users in Japan is projected to reach 51.6 million by the end of 2029, up 140% over five years, according to Japan's ICT Research and Consulting. The technology's spread is complicating risk management.

In 2021, U.S. real estate information company Zillow Group recorded a quarterly loss of more than $300 million after its AI-powered home-buying and resale business misjudged housing market fluctuations. The AI failed to capture pandemic-era shifts in housing prices.

In 2024, Air Canada faced compensation claims after a chatbot provided customers with incorrect information regarding airfare discounts.

According to a study published in 2025 by KPMG and the University of Melbourne, 56% of workers who use AI have made work-related mistakes caused by the technology.

"Insufficient AI governance increases the risk of lawsuits and customer defections, which can negatively affect a company's cash flow and growth potential," said Jun Okamoto, an executive partner at KPMG.