Even the Pope is weighing in on the dangers of unchecked deployment of artificial intelligence (AI). The pontiff’s Magnifica Humanitas presents risks in a theological context, but they translate easily into corporate action points: businesses must consider the implications in terms of treatment of customers, displacement of workers, accountability in decision-making, and exposure to all-powerful big tech.

The timing and nature of the publication – Anthropic co-founder Chris Olah was present at the Vatican City last month where he spoke out in favour of broader stakeholder oversight – is telling. It coincides with a shift in public sentiment on AI, most visible in anxiety about job security, opposition to power and water-hungry data centres, and loud indictments of governance lapses.

Each pain point is backed up by abundant data points. The World Economic Forum estimates that, even as job numbers grow through 2030, nearly 60% of the global workforce will need to be upskilled or reskilled. The International Energy Agency projects that global data centre electricity consumption could more than double between 2024 and 2030, with emissions by hyperscalers already soaring.

A recent review conducted by ISS STOXX found that, of more than 3,000 US-listed companies, only 8% disclosed board-level oversight of AI and 9% acknowledged having established policies on AI.

For private equity, the combination of papal intervention and public spotlight might serve as an ESG (environment, social, governance) wake-up call. Investors are highly attuned to the commercial consequences of AI, slicing up industries into those likely to benefit from efficiencies and those set to be torn apart by automation. When it comes to responsible AI policies, progress is more of a mixed bag.

“Part of the problem is it’s a big, nebulous topic,” said Peter Dunbar, a principal in the responsible investing team at StepStone Group. “When we first asked managers about AI risk, the perception of risk was not moving fast enough. The most common response was, ‘Don’t worry, we are deploying it in our portfolio companies as quickly as we can.’ They were getting the wrong end of the stick.”

Getting formal?

While global sponsors are ploughing resources into AI and entering alliances with the likes of Anthropic, OpenAI and Google Cloud, the broader picture in Asia is one of experimentation. Private equity firms of different sizes are applying AI systems internally, assessing what works and what doesn’t, and encouraging portfolio companies to do the same.

The emphasis on exploration almost regardless of cost – because the potential long-term efficiency gains could be so substantial – reflects a lack of joined-up thinking on ESG, according to Steve Okun, founder of APAC Advisors, a consultancy specialising in sustainability and stakeholder engagement.

“People think AI is going to be like air, available for as long as they want and affordable enough for them to consume as much as they need. You ask what their budgets are and it’s somewhere in between ‘We don’t have one’ and ‘We don’t know,’” he said.

“But what if electricity prices go up, data centre capacity is limited, or you have a single-source issue, and the cost of AI in three years is 5x what it is today? If there is a possibility of this happening, should companies be rolling back planned job cuts or recruiting associate classes at 80% as opposed to 50%?”

That kind of scenario analysis is indicative of an ESG mindset that Okun believes is seldom applied to AI by Asia’s private equity community. In his view, not enough managers have sought to define an approach to responsible investment whereby technology adoption is accompanied by risk mitigation.

Dunbar observed that some GPs globally are beginning to introduce formal AI policies that go beyond the regulatory-driven strictures of the 2024 EU Artificial Intelligence Act. In some cases, AI governance committees are being established to oversee compliance.

Governance has emerged as an initial focal point, in part because cybersecurity and data security risks are relatively well understood. It helps that the US National Institute of Standards and Technology (NIST) and the International Organisation for Standardization (ISO) have well-established systems covering best practice and certification, respectively.

Appropriate use of AI is also top of mind amid two ongoing lawsuits involving recruitment software platforms. Workday stands accused of running an AI-enabled screening tool that systematically disadvantaged job applicants based on age, while Eightfold is said to have ranked candidates using information from public databases and social media without their consent.

Australia-based Allegro Funds introduced an ethical use of AI policy last year having been questioned on the topic by three of its 10 largest LPs, according to Menno Veeneklaas, a strategy and ESG operating partner at the firm. The policy was rolled out internally and then shared with portfolio companies, which were asked to use it as a template for their own efforts.

Navis Capital Partners has done much the same, concentrating on when and how AI tools are used. “We don’t want to have agentic AI running wild in a company and people not doing any work, but it’s more about how data is used – what is limited to tools that we license and what can go onto external free and freemium models,” said Bence Szegedi, a senior director for ESG portfolio operations at the firm.

Future of work

Navis and its peers have barely scratched the surface on labour issues but encouraging portfolio company staff to test how AI might transform their roles is considered a good starting point.

A Boston Consulting Group study published this year breaks down roles based on disruption – those that will be amplified, rebalanced, divergent, substituted or enabled by AI. Lawyers are likely to be amplified because AI augments human capabilities and expands demand. Lab technicians will be enabled as AI reshapes how tasks are performed without fundamentally altering how work is structured.

The study found that 50%-55% of jobs across all six segments will be reshaped, not removed, while 10%-15% are vulnerable to elimination. Those tend to fall into the substituted and divergent segments: call centre representatives, certain financial analysts, insurance sales agents, and IT support technicians.

As private equity firms pursue AI-driven operational efficiencies, attention naturally turns to businesses with workforces that skew towards substitution and divergence. This does not necessarily mean wholesale redundancies. Affinity Equity Partners trialled AI agents at Ubase, a South Korea-based call centre and telemarketing services provider, but implementation is expected to be limited.

“You have to ask at what point does a customer prefer to speak to a real person rather than AI,” said Sarah Pang, an executive director and head of ESG and sustainability at Affinity. “If you want to reschedule your debt into six payments, the AI agent needs to check with the manager. At some point, a human must make a judgement call. It’s not as straightforward as getting rid of everyone and using AI.”

Allegro went through this process with call centre-based debt collection business Symbos, concluding that basic workflows could be automated and this would enable staff to focus on more complex tasks. Veeneklaas observed that relatively few jobs are suitable for 100% automation, citing coding and highly repetitive data entry as examples.

The GP has only seen three meaningful substitution events across its portfolio. Government consulting business Scyne Advisory let go a sizeable portion of its staff deemed ill-equipped to handle an increasingly digital-centric workloads – but replaced them with AI-native humans. Meanwhile, both Camp Australia and Discovery Parks have pared their locally based software development teams.

“Camp Australia [an out-of-school care business] had 20 devs on about AUD 200,000 each per year, which is a significant overhead for a mid-size company,” said Veeneklaas. “A Manila-based dev costs AUD 40,000 per year, while Claude Code is about USD 1,000 per year and it can do 5x more work.”

At the same time, another Allegro portfolio company, law firm Slater & Gordon, is rolling out an AI-enabled ERP platform that is expected to drive business growth without hiring additional lawyers.

This notion of deferred recruitment or employee attrition is highlighted by multiple private equity firms – and there is a social cost if handled insensitively. Much of the focus is on limited opportunities for young people entering the workforce, but there are gender equality issues as well given women are more likely than men to occupy roles vulnerable to AI.

“The US was the first to really embrace AI and that’s where the impact is being felt with people losing jobs and companies not hiring because they’re putting budget into AI infrastructure instead. That’s what leads to a political backlash, and it will spread everywhere,” said Okun of APAC Advisors.

“If businesses don’t recognise this, it will make the backlash so much worse. If you knew a backlash was coming one year from now, would you want a press release out there saying you’re all in on AI? It doesn’t mean you’re not all in, but you might communicate it differently.”

Getting granular

No one disputes that the US is ground-zero for AI fallout, but there is some expectation of geographic variation. Jason Corsello, founder and general partner at Acadian Ventures, which invests in work-related technologies, is a fierce advocate for AI having a net-positive impact on job creation. However, he accepts “there will be side effects” and is producing an impact report to get ahead of them.

In contrast, Frankie Fang, a founding managing partner at China-focused fund-of-funds Starquest Capital, notes that his portfolio GPs barely mention AI in an ESG context. As an energy transition-focused investor, Impax Asset Management is asking portfolio companies – in China and across Asia – to think about responsible AI, but it doesn’t buy into the notion of US-style backlashes.

“There are a lot of manufacturing businesses in Asia that are heavy on the human resources side, so governments will be careful about controlling the pace of AI adoption,” said Nana Li, head of sustainability and stewardship in Asia Pacific at Impax. “Then you have countries like Japan and South Korea, where populations are ageing. AI helps address that supply side problem.”

Regardless of local nuances in terms of execution, investors across the region anticipate a swift escalation in how AI features in ESG reporting, as opportunity and risk.

There is already evidence of LPs asking questions that extend beyond governance risks. Norges Bank Investment Management, which is responsible for Norway’s Government Pension Fund, has outlined broad areas it will focus on in assessing responsible AI credentials. These include inequality and discrimination risk as well as long-term effects on human capital management.

At a recent meeting with a development finance institution (DFI), Szegedi of Navis was informed that exposure to companies undergoing unexpected mass layoffs had been added to the organisation’s list of investment restrictions. He interpreted it as a reference to AI-related disruption. “It’s not surprising in the light of what has been happening recently,” Szegedi added.

This seems exceptional partly because many LPs are described as held back by a reluctance to add to already sizeable disclosure requirements or a lack of familiarity with an area that is new, fast-evolving and in certain parts highly technical. The Principles for Responsible Investment (PRI) is said to be looking to fill that knowledge gap with a checklist of AI questions for LPs.

In a March blog post, Thomas Abrams, the organisation’s head of human rights, social and governance issues, listed actions investors may take now, citing existing frameworks developed by the likes of the Organisation for Economic Cooperation and Development (OECD) and the UN Educational, Scientific and Cultural Organization (UNESCO).

These include finding out which AI systems companies are deploying; clarifying board oversight for AI use; assessing how AI risks are identified and monitored; checking regulatory preparedness; reviewing protocols for assessing impacts on workers, data subjects, and the environment; seeking transparency on AI use cases, safeguards, and reporting; and ensuring proper use of third-party technology providers.

“While even the medium-term impact of [AI] changes remains nebulous, and the technology may feel complex or unfamiliar, investors can still take meaningful action to mitigate risk whilst realising opportunities,” Abrams observed.

Data dynamics

Inevitalby, quality of data is a challenge, with industry participants resigned to relying on qualitative information. Some are working on internal fixes. Jun Tsusaka, CEO of Japan-focused NSSK claims the four AI modules he has in development “will create the data infrastructure needed to move from qualitative narrative to quantitative, comparable indicators across the portfolio.”

This does not address holes in external datasets, which are numerous. Ovidiu Patrascu, a senior director for responsible investment strategy at Nuveen, noted earlier this year that the power usage effectiveness (PUE) metric frequently relied on as a measure of data centre sustainability does not factor in the source of power. He advocated using a broader lens to assess environmental performance.

However, extracting any material information from technology providers can be problematic. Anthropic does not disclose to Allegro hourly power and water consumption of Claude Pro; the firm only knows charge-per-seat and overruns, according to Veeneklaas. The relative resource consumption of prompt-heavy usage versus treating Claude as a version of Google search is a black box.

“If this is a massive environmental issue, we will need to get our heads around how we quantify it and report on it at the holding company and portfolio company levels. We’ve done it on climate for scope one, two, and three emissions, so we must do the same for AI usage,” he added.

“It’s possible that some more progressive countries will end up putting some impost on that. If you’re running a data centre in their country and it’s being used for heavy-duty AI work, maybe all that carbon has to be offset by planting trees.”

StepStone’s Dunbar draws on the climate comparison as well, both to emphasize the unavoidability of the issue – “If you are a universal asset owner, this isn’t something you can diversify away from” – and to identify potential entry points in terms of data collection. Investors are already thinking about tools used for climate change that could be applied to AI assessment. In this context, it is acceptable to start small.

“There are basic metrics like the percentage of portfolio companies with responsible AI policies. Scenario analysis is also a good idea. With climate, you look at different scenarios from 2 degrees [Celsius] to 8.5 degrees and there could be an equivalent way of doing it for AI,” he said.

The uncertainties around AI readily pile up, while attempts to capture its ESG impact are routinely outpaced by technological development. The NIST and ISO risk management frameworks are barely three years old, yet they were conceived as responses to generative AI. Agentic AI’s breakthrough moment was less than 18 months ago.

But the speed of a technology’s rise and the breadth and complexity of its impact should not deter investors, especially when baseline approaches to ESG risk assessment are well-embedded.

“Do we know how it’s going to play out? Of course not. Is it too soon to know that AI is going to become restricted? Yes. However, it’s not too soon to be thinking about it from a traditional ESG perspective,” said Okun of APAC Advisors. “If you have responsible investment policies around business ethics, sourcing, the environment, and employee health and safety, why wouldn’t you have one for AI?”