Union contracts are becoming HR’s AI governance playbook
SHARE THIS ARTICLE
HR leaders, both in the United States and across the globe, should pay attention to an unexpected source of workplace AI policy: collective bargaining agreements, writes Dr Gleb Tsipursky.
A July 2026 Axios review found that the NewsGuild-CWA alone has roughly 85 to 90 contracts with explicit AI provisions, making union agreements an increasingly important form of practical AI governance. The provisions vary, but many require some combination of notice, bargaining, consent, limits on replacement, and enforceable safeguards. For HR, the lesson reaches far beyond unionised workplaces. These contracts show what employee participation looks like when organisations must define it before deployment rather than improvise it after resistance erupts.
Gallup reported in April 2026 that half of employed Americans use AI in their role at least occasionally, while employees at organisations that have adopted AI report more disruption than employees elsewhere. Productivity gains often appear at the task level before organisations redesign work around the technology. That gap turns AI change management into a core HR responsibility. When leaders buy tools faster than they redesign jobs, expectations, training, and decision rights, employees absorb the ambiguity.
Union contracts force management to answer practical questions before rollout. What work will the system perform? Which decisions stay with people? What employee data will it collect? Who can challenge an error? How will productivity gains affect staffing, workload, pay, and development? The US Department of Labor’s 2024 employer best practices similarly emphasised worker voice, transparency, human oversight, training, and worker-data protection. HR can use that structure even where collective bargaining never enters the picture.
The Washington-Baltimore News Guild’s dispute with Politico shows why those questions matter. The Communications Workers of America (CWA) reported that an arbitrator found Politico had violated its collective bargaining agreement after introducing AI tools that bypassed negotiated safeguards, and management later shut down two tools associated with inaccurate content. The HR lesson concerns human oversight, role clarity, and escalation rights. Employees closest to quality failures often see problems before senior leaders do. A governance process that gives them no meaningful way to stop or challenge a flawed system can turn a manageable pilot problem into an enterprise problem.
ZeniMax workers reached a different kind of agreement with Microsoft. CWA said the agreement committed the company to AI uses intended to augment human capabilities and required notice when implementation could affect bargaining-unit work, with bargaining over those impacts available upon request. That approach treats employee participation as part of operational design rather than a communications exercise. HR leaders can borrow the same workplace AI adoption logic by involving affected employees before leaders lock in workflows, performance expectations, or staffing assumptions.
SAG-AFTRA’s 2023 TV and theatrical agreements offer another model. The union negotiated requirements around informed consent, notice, compensation, and bargaining for certain uses of digital replicas and synthetic performers. Most companies will never face the same likeness and intellectual-property issues, but the employee relations principle transfers easily: when AI materially changes what employees contribute, how management evaluates that contribution, or whether the organisation still needs the role, HR should define the rules before the technology creates a dispute.
Most American workers lack these protections through a union contract today. The Bureau of Labor Statistics reported that 16.5 million wage and salary workers, or 11.2 per cent, were represented by a union in 2025. For HR, that workforce policy gap creates an opportunity to build credible participation into AI deployment before employees demand formal protections.
The first discipline should be advance notice tied to a job-impact assessment. Before a material deployment, HR should document which tasks will change, which roles face greater or lesser demand, what new skills employees will need, what data the system will touch, and which decisions require human review. NIST’s voluntary framework encourages organisations to manage responsible AI adoption across design, deployment, use, testing, and evaluation rather than treating risk review as a one-time approval. HR should translate that life cycle logic into a people-impact review that begins before implementation and continues after launch.
Second, HR should create representative design groups with real influence. Include frontline employees, managers, technical specialists, legal and security staff, and people whose work will change most. Give the group authority to test assumptions, recommend workflow changes, and flag uses that require escalation. Participation without influence quickly becomes theatre.
Third, define boundaries before employees encounter them in practice. HR should specify prohibited uses, required approvals, appeal rights, documentation requirements, monitoring limits, and conditions for pausing a system. This matters especially when AI touches hiring, promotion, evaluation, discipline, or termination. The EEOC’s current enforcement plan specifically identifies the use of AI and machine learning in recruitment and hiring as a potential source of unlawful barriers, which gives HR AI governance a direct compliance dimension alongside the change-management one. EEOC resources also warn that software, algorithms, and AI used to assess applicants and employees can raise AI employment risk under disability law.
Fourth, connect productivity gains to an explicit workforce plan. HR should make leaders answer what happens when AI saves time. Will teams handle more volume, improve service, reduce overtime, eliminate low-value tasks, retrain employees, redeploy people, or reduce headcount? The World Economic Forum’s Future of Jobs Report 2025 found that employers widely expect both automation and major skill shifts, while upskilling remains the most common planned workforce response. That makes workforce planning inseparable from AI strategy. Employees can handle difficult news more effectively when leaders explain the trade-offs early and provide credible transition paths.
Fifth, build enforcement and measurement into the process. Employees need a channel to report failures without fear of retaliation, leaders need named responsibility for corrective action, and major deployments need scheduled reassessment. HR should track adoption, training completion, error reports, workload distribution, appeals, service quality, and deployments changed after employee feedback. Those metrics show whether participation improves decisions or merely produces meetings, and they give the CHRO evidence for the executive team and board.
Some executives will argue that these processes slow innovation. Poorly designed governance can. Yet purchasing speed tells HR little about speed to value. Gallup’s 2026 findings show substantial AI-related disruption while broader workflow transformation remains uneven. The practical question is which process produces durable adoption: one that treats employee concerns as friction, or one that uses those concerns as information about job design, training, risk, and implementation.
That distinction should put HR near the centre of enterprise AI decisions. Technology teams can evaluate capabilities, security teams can assess exposure, and legal teams can interpret obligations. HR brings the workforce system into the room: jobs, skills, incentives, performance, employee relations, mobility, communication, and trust. An effective AI adoption strategy needs all of those elements working together.
Union contracts are showing HR what structured employee influence looks like under pressure. Companies without unions can still adopt the strongest features voluntarily: advance notice, representative participation, clear boundaries, transparent workforce consequences, appeals, enforcement, and measurement. HR leaders who build those mechanisms before conflict emerges can reduce resistance while improving the quality of AI decisions. The bargaining table offers a useful warning and a useful blueprint. Employees will seek a voice in how AI changes their work. HR can design that voice before employees conclude that formal bargaining provides the only reliable way to get it.
Dr Gleb Tsipursky is the CEO of the future-of-work consultancy, Disaster Avoidance Experts. This piece has been adapted from The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026).
Want to see more stories from trusted news sources?Make HR Leader a preferred news source on Google.
