Teachers, nurses, university employees and public servants saving for retirement appear to be unknowingly funding technologies that could be deployed to target or surveil civilians in conflict zones. According to new analysis conducted by Empower in partnership with the Business and Human Rights Centre (BHRC), Open MIC and Heartland Initiative, at least 182 private and public pension funds, from various countries, have invested in companies developing high-risk artificial intelligence (AI) systems. 

Our research identified 32 companies – including OpenAI, Anthropic and xAI – developing generative AI systems being developed, marketed or positioned for current or future military applications. We analysed their investors to assess whether pension fund managers are conducting adequate human rights due diligence on the companies they are funding. Pension fund managers and pensioners may be unaware of this hidden risk festering within their portfolios. Because technology companies’ policies on responsible AI have rapidly devolved – and given many generative AI startups are privately held and beyond traditional shareholder oversight – current ESG screens and due diligence processes are outdated and insufficient to detect or prevent investment in the weaponisation of AI. This means even responsible investors are likely exposed to investments in what some are calling “supply chain risks” and what others are calling “the erosion of international law”.

Amidst the fast-paced fusion of commercial AI development and military applications in a largely unregulated space, responsible investors must navigate these challenges creatively and take action to ensure their investments align with pensioner values and rights-based risk management.

This briefing outlines:

• the risks pension funds face from investing in AI-powered defence tech;
• how these risks have already materialised;
• which funds were reportedly linked to these technologies; and
• actionable solutions pension fund managers can implement now to reduce their risk.

Generative AI is a critical component of new kinds of controversial weapons

The rapid rise of generative AI intertwining with military applications has resulted in a new, high-risk investment arena. Previously civilian-oriented tools – from AI surveillance systems to generative models like ChatGPT – are increasingly being repurposed for military use. Now that this trend has intensified, investing in certain technology companies has become indistinguishable from investing in war, requiring more and better due diligence to underpin investment decisions.

Unlike many traditional AI systems which assign labels or predict predefined values, generative AI produces new content, making its outputs inherently less predictable. Military applications of generative AI are particularly vulnerable to bias, hallucinations and misalignment – risks which are especially dangerous when applied to targeting decisions. This means that generative AI used for military surveillance, command support and targeting could result in indiscriminate effects (i.e. by not adequately distinguishing between combatants and civilians), could cause large-scale harm (i.e. many people are impacted), and there are a number of unresolved ethical and legal questions (particularly concerning how to define human agency and accountability for decisions). Applying generative AI to military operations challenges accepted norms of warfare, human rights standards and international law.

What are the main concerns regarding generative AI-powered warfare?

1. Generative AI poses significant reliability risks, as it can produce confident but incorrect or fabricated outputs.
2. Generative AI is used to accelerate military decision making, raising concerns about unwarranted or reckless escalation and lack of accountability for harms.
3. When training foundation models relies on the mass surveillance of civilians, and those datasets are integrated into military operations, civilian data is transformed into military infrastructure – without public awareness, democratic oversight or individual consent.

For more information about why this is concerning, and answers to the following questions, visit the FAQ page:

• Is generative AI already being deployed in conflict zones?
• Why is deploying generative AI in conflict-affected areas high risk?
• Which are the 32 companies we analysed that are designing, developing, deploying, marketing or otherwise positioning generative AI as a tool for militaries?

Pension funds are left exposed to heightened risk in increasingly unstable times

Defence stocks and AI firms with military applications have shown strong short-term performance, driven by surging military spending, heightened geopolitical tensions and post-crisis market rallies. For short-term investors, this rally has been lucrative. However, for long-horizon, value-oriented investors such as pension funds, these gains mask a deeper material risk profile that short-term returns cannot offset.

AI systems used in targeting, surveillance or autonomous decision making in conflict settings raise serious legal questions around accountability, proportionality and civilian harm. Companies developing or deploying such systems face growing risks of litigation, sanctions or criminal complaints linked to states with universal jurisdiction, or even investigations by the International Criminal Court (see: the Office of the Prosecutor’s latest guidelines).

Pension funds face material reputational risks when investments fall out of step with beneficiaries' values. Exclusionary screens have long existed precisely because the intangible cost of being associated with certain products, such as weapons, is too high. Public campaigns and member activism demonstrate this clearly: pensioners and civil society in countries such as the United Kingdom, Canada, United States, Netherlands, Sweden and Denmark have repeatedly pushed funds to divest from companies linked to conflicts in Myanmar, Yemen, Sudan and Palestine, among others.

Major pension funds in Norway and Switzerland have divested from companies producing controversial weapons and have now included technology companies that service the defence industry, such as Palantir. Even smaller funds, like Tower Hamlets Council in London, have acted to divest from companies implicated in alleged war crimes. “[It’s] no surprise that people who dedicate their lives to providing local government and education support services don’t want to see their pension pots invested in death, destruction and mutilation,” said Kerie Anne, from Tower Hamlets. “Swedish pension money should not be invested in companies that sell weapons to or fund the Myanmar military junta,” noted Yadanar Maung, a spokesperson for Justice For Myanmar who has been organising around AP7’s decision to exclude four companies linked to the Myanmar military.

How are pensioners linked to high-risk AI technologies?

In the chain of private capital investment, there are several key types of actors. Venture capital (VC) or private equity firms, referred to here as general partners (GPs), are the entities in a partnership structure that manage and make investment decisions for a fund but do not themselves provide most of the capital invested (GPs usually contribute less than five per cent of the money in a given fund). Meanwhile, the limited partners (LPs) are generally institutional investors that collectively provide the bulk of the capital as passive investors in a fund. Examples of LPs include: private pension firms, public pension firms, sovereign wealth funds and high net worth individuals, among others.

Given the drastic changes in policy and practice of many technology companies, existing exclusionary screens, due diligence processes and ESG assessments intended to help LP investors avoid funding weapons and conflict are today insufficient. Given that many of these 32 companies we researched are privately held, some of the traditional tools for investors to enhance corporate oversight (such as filing shareholder proposals) are not applicable. These outdated processes likely cannot detect when AI is being weaponised. Thus, even responsible investors are not immune to contributing to AI weapons development.

Without a concerted effort by LPs to fill this gap, GPs prioritising short-term profits risk investing pensioners’ savings in technologies that are likely to exacerbate conflict.

Which GPs are investing in these 32 companies? Find out more on our FAQ page

Strong returns do not require funding needlessly dangerous technologies

Investment has surged in two key areas: defence and artificial intelligence. Analysts are projecting global AI spending to reach nearly USD 2.5trillion in 2026, a 44% increase from 2025. At the same time, global military expenditures have climbed to USD 2.7trillion – the highest in three decades – prompting UN Secretary General António Guterres to observe that “the world is spending far more on waging war than in building peace”. That amount is roughly 750 times the United Nations’ operating budget. This disproportionate focus on AI and defence diverts resources from social programmes, poverty reduction, education, health, peacebuilding, diplomatic work, environmental protection and critical infrastructure.

According to the IMF, every USD1 spent on conflict prevention averts up to USD103 in future conflict costs. Investors have the power to drive change while pursuing high-yield strategies – but this requires a deliberate shift away from funding unchecked and unregulated military applications of technology. This requires investing in companies that are committed to respecting international law and understand their responsibility to avoid contributing to conflict and civilian harm.

Pension funds, as "universal investors", have a fiduciary responsibility to protect the stable economic and social conditions that their members depend on, guided by a longer-term outlook to align financial performance with societal well-being. Under the UN Guiding Principles on Business and Human Rights, funds have both the power and responsibility to steer capital toward safety and rights-respecting innovation.

To the LPs that are worried about missing out on investing in specific GP funds:

Portfolio-wide risk outweighs single-fund access. If a GP refuses risk transparency or contractual safeguards that are more fit-for-purpose in the current context, that itself is a red flag. Reputational harm affecting the entire fund, especially public funds, is far more damaging than losing access to a single GP. If multiple LPs request similar safeguards, norms shift. Individual LPs often underestimate their leverage and influence within the larger LP ecosystem. We aren’t going to get anywhere if everyone is waiting for someone else to take the first step.

To the LPs that are concerned about overstepping in a 'limited’ partnership: 

Risk management is not control. Requiring baseline risk protections (such as reporting requirements, side letters, ESG or human rights policies) does not convert an LP into a controlling party. LPs routinely negotiate fee structures, set compliance standards and impose exclusion lists. Adding human rights risk oversight concerning technologies that pose the highest risk to democratic systems and geopolitical stability is simply an extension of prudent risk management for a pension fund's essential mandate. As global standards evolve in step with the recognition of ballooning risk, inaction may become harder to defend than reasonable, documented oversight.

To the LPs that see this as critical to address, but are wary of making a misstep:

This is about mitigating risk, not achieving perfection. Concrete first steps can be simple and bounded, such as joining a working group of investors who are actively debating the recommendations listed below. Even light-touch, standardised safeguards can have portfolio-wide protective effects. When beneficiaries, board members or the media inquire about concerns, it is far better to respond that steps are being taken rather than being caught flat-footed.

To summarise, pension funds can pursue strong returns without backing AI that enables war crimes.

There is no conflict between fiduciary duty and responsible investment. Major pension funds have already implemented delimited exclusions or divested from controversial weapons on human rights and reputational grounds, proving that protecting beneficiaries' values and protecting their returns aren't mutually exclusive.

These decisions didn't happen in a vacuum – they followed sustained campaigns by beneficiaries demanding alignment between retirement savings and values. They aren’t top-down ESG initiatives. They reflect bottom-up demand for accountability from the teachers, nurses and public servants demanding their money is not funding harm.

Pension funds should:

1. Tie capital commitments to data governance safeguards that force the decoupling of civilian and military data. Capital commitments should be linked to robust data protection, privacy and consent frameworks, helping to ensure that civilian technologies are not repurposed for militarisation.
2. Require human rights and humanitarian law risk disclosure from GPs and big tech companies. Make regular reporting on potential military or dual-use applications and human rights impact assessments a condition of generative AI investment.
3. Embed human rights due diligence in GP contracts and processes. Set expectations that GPs integrate human rights and humanitarian law considerations into startup valuation, investment evaluation, monitoring and reporting.
4. Draw red lines, abstain from high-risk AI investments, and ensure that GPs are actively mitigating risks to civilians in their investment decisions. Establish clear boundaries for AI defence applications that trigger divestment. Revise existing exclusionary screens to explicitly rule out investments in high-risk generative AI technologies that may reasonably cause, contribute to or facilitate IHRL or IHL violations.
5. Engage publicly and collaboratively to reshape market norms around responsible AI toward deescalation and peacebuilding. Large pension funds in particular should use their influence to signal changes in expectations for tech investment to the broader investor ecosystem.

Accelerated timelines in an already hyper-hurried AI industry have left little time for the integration of international law or other key safety or security considerations that high-risk technologies should require. But market incentives, like war itself, are human-made. Pension funds have the power to choose a path of facilitating de-escalation over reckless deployment.

Cover image credit: Shutterstock