The question I put to a union pension fund CIO was simple, and it made him uncomfortable: has he made a deliberate decision not to invest in private equity and venture capital funds backing the development and deployment of worker-replacing AI systems? That union plans have already invested in such funds is hardly in doubt and the illiquid, penalty-laden terms governing these private investments make an exit difficult or effectively impossible. So, I set aside existing allocations and framed the question prospectively: Has he made a deliberate decision not to allocate to funds backing the companies building agentic AI systems designed to displace the very workers whose retirement security he was appointed to protect?
This is not an ESG question. Union plans are not ordinary institutional investors. Their capital is held in trust for workers, and is governed, in many cases, by investment criteria that explicitly support and protect workers' interests.
These criteria are not aspirational statements. They are the standards that trustees have formally agreed to apply. For example, the Service Employees International Union (SEIU), which has more than 2 million public and private sector members in the United States and Canada, with combined pension assets of more than $1 trillion, states it “seeks to make certain that our members' pension funds are managed in a manner consistent with their long-term economic interests. We consider this our primary goal, and one that is consistent with encouraging responsible investment and proxy voting policies, active ownership and responsible business behavior by companies our funds invest in.”
Nothing in this argument asks union plans to sacrifice risk-adjusted returns. SEIU, the AFL-CIO, and others have long demonstrated that labor-aligned investment criteria can be applied within a rigorous fiduciary process. The question is not returns-versus-workers. The question is whether plans with explicit labor-friendly investment policies can justify allocating capital to the companies and funds driving the displacement of the very workers those plans exist to serve.
This is not a hypothetical. AI is already displacing workers and locking out many seeking entry-level positions, and that is likely to continue. Block, for example, cut 40 percent of its workforce — 4,000 people — as it deployed AI tools that CEO Jack Dorsey says "fundamentally change what it means to build and run a company. … Within the next year, I believe the majority of companies will reach the same conclusion."
Many unions and their supporters agree with Dorsey that AI will dramatically transform the workforce, causing massive job displacement for union members across the occupational spectrum, ranging from bartenders and security guards to assembly line workers across entire industries.
AFL-CIO President Liz Shuler recently called AI the "single biggest threat to working people of our lifetime," warning that employers are deploying it without adequate oversight, while Randi Weingarten, the president of the American Federation of Teachers, said, “If we don’t figure out how to harness [AI in contracts] and how to have educators lead it, what’s going to happen is it’s not just going to be about some job loss. It’s really going to be about the machine, basically taking over from human beings.”
However, with lax federal oversight of AI, unions are being pushed to negotiate safeguards on a contract-by-contract basis, while the deployment of this technology moves faster than any bargaining calendar can keep up with.
Predicting the rate, timing, and location of job displacement by generative AI is a challenge for even the most attentive journalist or economist.
The fact that methodological flaws beset current methods for determining whether a job or industry may experience AI-driven turbulence has not stopped people from making their own predictions. Some forecast a dire impact on workers: Anthropic CEO Dario Amodei projects that half of all entry-level white-collar jobs could disappear within five years, with unemployment rising to 10 to 20 percent. Investor Vinod Khosla predicts AI will replace 80 percent of jobs by 2030. Roman Yampolskiy, an AI safety researcher, goes further, projecting unemployment levels "we've never seen before — not 10 percent, which is scary, but 99 percent." (Coinciding with Anthropic's anticipated IPO, Amodei has walked back his more alarming predictions, now framing AI as an engine of productivity rather than a catalyst for mass unemployment.)
Others are more sanguine. Princeton's Arvind Narayanan and Sayash Kapoor argue that AI will follow the same diffusion arc as electricity and the internet, unfolding over decades rather than years. Past general-purpose technologies, they note, took that long to fully impact employment. AI should be expected to do the same.
Historical analogies are instructive but ultimately insufficient: AI’s broad applicability means its effect will be felt across every sector at once, including those most heavily represented in institutional portfolios.
The timeline debate is legitimate. The direction is not — it moves with the force of Hegelian determinism.
Every optimistic historical analogy rests on the assumption that workers can retrain into new roles before those roles are also automated. OpenAI’s CEO Sam Altman put it plainly: "We'll find new kinds of jobs, as we do with every tech revolution."
The Carnegie Endowment for International Peace examined the historical record on retraining and found it does not support Altman's confidence. Past U.S. programs failed because they trained workers into skills that themselves became obsolete. AI makes this chronic problem acute. Unlike prior automation waves, AI is too fast, too broad, and too unpredictable to allow reliable program design. Carnegie finds meaningful returns only in STEM, technical, and health vocational training — precisely the fields where AI exposure is highest and climbing fastest. The retraining argument leads workers directly toward the jobs most likely to disappear next. AI learns new jobs faster than workers can be trained to perform them.
Retraining is not a solution to the future unemployment problem; it is a different version of the same problem.
And it extends beyond retraining to the promise of reemployment itself. Sam Altman captured the optimist's faith in a single sentence: "We'll find new kinds of jobs, as we do with every tech revolution." One frequently cited example of reemployment is that AI infrastructure--data centers in particular--will generate jobs and revitalize communities. For union pension funds, this argument has specific appeal: infrastructure projects built with union labor under project labor agreements have long been a cornerstone of labor-friendly capital deployment. Data centers serving hyperscalers are not.
A recent Harvard Gazette investigation examined the industry's central claim that new facilities bring lasting jobs to local communities and reached an unambiguous conclusion: "It's a significant false promise." Data center construction does create work, sometimes local and unionized, sometimes workers brought in from other states. But construction is temporary, typically lasting a year or two. What remains when the building is done is a facility employing between 20 and 50 people. A warehouse of servers, running in a community that was told it was getting an economic anchor.
Universal Basic Income is frequently proposed by AI proponents as a policy response to mass displacement. It is the wrong answer to the right question. UBI is a consolation prize for those whose lives are disrupted. It addresses income loss but not the social and psychological consequences of work itself (e.g., identity, structure, purpose, and community). It also does nothing to address inequality, normalizes displacement rather than challenging it, and makes techno-capitalism more tolerable while masking the symptoms of economic injustice without addressing the root causes of exploitation and inequality. UBI is not a response to exploitation and inequality; it is a mechanism for sustaining it.
In the 1970s and 1980s, labor unions attempted to use their financial clout to push back against companies engaged in aggressive anti-union tactics and the threat of moving jobs offshore. The results were mixed at best, though CalPERS emerged as a notable ally in pressing for responsible investment standards.
But this is a different challenge. Offshoring was a policy problem with policy remedies that the labor movement could contest in the political arena, and it affected specific, identifiable sectors. AI displacement is a broad-based technological condition that moves faster than politics, faster than regulation, and faster than any retraining program can follow. The tools available to trustees — investment policy, capital allocation, fiduciary scrutiny — are the same ones the labor movement has always had. What has changed is the stakes are higher and the consequences irreversible.
Union pension funds owe a fiduciary duty to retirees collecting today, to active members building toward retirement, and to future members whose careers and contributions the fund is implicitly counting on. AI is a known and documented threat to all three, on a timeline no one can fix with precision. But fiduciary duty requires prudence — the recognition that known risks demand serious action before the damage becomes irreversible. The risks here are known, and the data is already confirming the trajectory. Trustees who wait for the consequences to become undeniable before acting will find, when that moment arrives, that they have already failed the people they were appointed to protect.
One argument above all ends conversations about worker protection before they begin: China is investing heavily in AI as a national priority and ceding AI leadership would give China decisive advantages in economic productivity, military capability, and geopolitical influence. Any regulation, safety requirement, or social protection that slows U.S. AI development is, in effect, a gift to a strategic adversary.
The American Edge Project states it plainly: failure to lead in AI risks handing China "an edge that the United States will likely never recover."
The Trump administration has governed accordingly: revoking Biden-era AI safety oversight, releasing a deregulatory Action Plan framed around outcompeting China, issuing an executive order that prohibits states from adopting regulations that may be “barriers” to AI development, and committing the Pentagon to becoming an AI-first warfighting force.
The national security concern is legitimate. But it does not answer the question this essay is asking. Washington has chosen acceleration as its singular imperative and treated worker protection as a cost it cannot afford. That calculus may be defensible on national security grounds. For union pension fund trustees whose obligation runs directly to the workers absorbing that cost, accepting it without challenge is not stewardship.
The economic consequences of mass AI displacement are discernable: Workers are simultaneously producers and consumers. Displace them at scale, and you do not simply cut costs; you destroy the consumer base that sustains corporate revenues, erode the tax base that funds public services, and overwhelm a social safety net designed for cyclical downturns, not structural transformation.
The productivity gains from agentic AI flow upward to a small cohort of developers, investors, and platform owners who are already among the wealthiest people on earth. An unemployment crisis and an inequality crisis arrive together. Geoffrey Hinton, the Nobel laureate and godfather of AI, was characteristically direct: “Rich people are going to use AI to replace workers. It's going to create massive unemployment and a huge rise in profits. It will make a few people much richer and most people poorer.”
The conditions today are not identical, but the trajectory is recognizable. Anti-AI sentiment is already rising. An NBC survey found that AI had a net favorability rating among voters 18 to 34 of negative 44. A Gallup poll finds seven in ten Americans oppose constructing data centers in their local area, with nearly half strongly opposed. A Reuters/Ipsos survey found that seventy-one percent of respondents fear AI is causing permanent job loss. Workers are sabotaging their company's AI systems.
Researchers warn that these grievances are reviving dormant violent extremist movements and giving rise to new ones. "AI is becoming this driver of political violence, and that's a very new phenomenon," says Jordyn Abrams, a researcher at the Program on Extremism at George Washington University. Yannick Veilleux-Lepage, a political scientist at the Royal Military College of Canada, goes further, arguing that "AI generates the structural conditions historically associated with the onset of political violence,” conditions that, once established, have rarely remained contained.
The first incidents of politically motivated AI-related violence have already arrived. In April 2026, a 20-year-old allegedly firebombed OpenAI CEO Sam Altman's San Francisco home before proceeding to the company headquarters, where he told security staff he intended to set the building alight and kill anyone inside. Four days earlier, another attacker targeted an Indianapolis city councilman who approved a local data center project. Last month, two self-described "ecofascists" who carried out a deadly attack on a San Diego Mosque were motivated by both bigotry and an anti-technology extremist ideology.
At sufficient scale, a labor problem becomes a democratic one and a threat to domestic security. Recall the original Luddites, who smash the looms and wide frames that factory owners were using to automate their labor, were not against progress or anti-technology. They were opposed to the way the industrialists used technology to depress wages, erode workers’ rights, and force them into poverty, all to profit at their expense. Their protests succeeded at first. In 1812, Parliament sided with the industrialists, passing the Frame Breaking Act and making machine-breaking a capital crime. The state crushed the movement within a year. Scores were hanged. Many more were killed as owners took up arms against their own workers.
Dario Amodei has named the democratic stakes plainly: "The balance of power of democracy is premised on the average person having leverage through creating economic value. If that's not present, things become kind of scary. Inequality becomes scary. And I'm worried about it." Palantir CEO Alex Karp was more blunt: "The biggest challenge to AI in this country is political unrest. The country could blow up politically — and none of us are going to make any money when the country blows up."
Labor leaders have reached the same conclusion. AFL-CIO President Shuler has warned that if AI is left unregulated, without specific worker protections, and in the hands of asset managers and corporations seeking to maximize profits, "AI will increase economic inequality, curtail our rights, and undermine our democracy."
The question that remains--and that this essay has been building toward-- is whether union pension fund trustees and staff will act on that awareness or continue making private market investments that contradict everything the labor movement says it stands for.
After hearing the argument laid out above, the CIO's first reaction was concern about reputational risk and the discomfort of explaining to trustees that the plan holds positions in companies like Mercor ("The $10 Billion Startup Training AI to Replace the White-Collar Worker"), Artisan (whose tagline is "Stop Hiring Humans"), or Mechanize ("Our goal is to fully automate work"). That discomfort is understandable. It is also the wrong place to start.
His second reaction was more honest: "We're slitting our workers' throats with their own capital."
The plan's investment capital comes from workers. It is being deployed, in part, to fund the AI systems displacing those same workers — eliminating entry-level positions, automating labor-intensive roles, and financing infrastructure that generates profits for a narrow class of asset owners while eroding the contribution base union funds depend on. Fewer employed union members mean fewer contributions and a shrinking asset base. The technology displacing workers today is the same technology undermining the funding base that pays tomorrow's retirees. The solvency risk is not theoretical; it follows directly from the displacement math.
Fiduciary duty has never required trustees to wait for consequences to become undeniable before acting. It has required prudence and the recognition that known, documented risks demand a serious response before the damage is irreversible. The direction of AI displacement is not in dispute. And the stakes extend beyond the balance sheet: as this essay has argued, mass displacement at sufficient scale does not remain an economic problem. It becomes a democratic one, with consequences that no pension fund or society is positioned to absorb.
Union leadership is aware of the threat. The question is whether pension trustees and CIOs will translate awareness into investment decisions aligned with what the labor movement says it stands for. The time for that reckoning is now: trustees and professional investment staff must assess whether their future commitments to AI-driven private equity, venture capital, and data center infrastructure serve or betray the workers whose retirement security they were appointed to protect. The contradiction at the center of the portfolio is no longer abstract. It has a price, and the workers are paying it.
Angelo Calvello is the founder of C/79 Consulting, a columnist for Institutional Investor ("The Dissident"), and the host of Against Consensus. He serves as a trustee on the Woodridge Police Pension Fund and chairs the Climate Advisory Panel at the Maryland State Retirement System.
