Unions are giving US public employees a real say in how governments adopt AI. Contract protections that stop automated systems from deciding dismissals or discipline are, for now, a more direct route than waiting for legislation.
Key takeaways
- Municipal workers in Oakland, California, have authorised a strike partly over how the city might deploy artificial intelligence, with 94.1% voting in favour, according to the Guardian.
- The main worries are increased surveillance, inaccurate automated performance evaluations and the risk of being laid off or disciplined because of an AI system.
- Public-sector unions have unusual leverage over AI adoption because the employer is a government that must negotiate working conditions with them.
- A strike authorisation shows that members are organised, but it does not show that protections have been won, and the final terms of any agreement are not yet known.
The argument: bargaining tables are where AI rules are being written
AI tools are entering the workplace faster than laws to govern them. In the United States there is no single national framework for AI in employment, so the practical rules are often set at a lower level: by individual employers, in procurement decisions, and increasingly in labour contracts.
This article argues that collective bargaining, especially in the public sector, is currently the most effective way for workers to influence how AI is used on them. Legislation and voluntary corporate principles also matter, but they are slower, broader and harder for an individual worker to enforce. A union contract is specific, enforceable through grievance procedures, and negotiated by the people who will live with the result.
The Guardian reports that, as governments adopt AI, workers are gaining ground over how it is used and securing safeguards against its harms. That framing is more optimistic than the evidence so far supports, as discussed below. The direction of travel, however, is clear: AI is becoming a standard bargaining issue, much like pay, hours and health and safety.
The demands are mostly about process rather than outright bans. Workers are not generally asking governments to give up AI. They are asking that a human remain responsible for consequential decisions, that monitoring be limited and disclosed, and that automation not be used as a quiet route to job cuts. Demands of this kind fit naturally into a contract, which is part of why bargaining suits them.
Evidence one: the Oakland strike vote shows workers will act over AI
The clearest recent signal comes from Oakland, a city close to Silicon Valley. The Guardian reports that the union representing a group of city employees voted on 22 October to authorise a strike, with 94.1% in favour. Among the issues behind the vote is a demand for protections ensuring that workers are not laid off or disciplined by AI.
The size of the majority matters. Strike authorisations are a standard negotiating tool, and a strong vote tells the employer that members are prepared to withdraw their labour if talks fail. That AI-related protections are part of the dispute shows that workers no longer treat the technology as a distant or abstract concern. They see it as something that could shape their jobs within the term of their next contract.
The concerns reported are concrete. City staff worry that AI systems could increase surveillance in the workplace and judge their performance inaccurately. Both worries reflect known weaknesses of automated management tools. Monitoring software can gather far more data about workers than a human supervisor would, and evaluation systems can produce scores that are hard to challenge when the reasoning behind them is opaque.
According to the Guardian, one city employee, a spatial data administrator and union member, described the aim as getting the city administration to be deliberate about how it engages with AI products. That is a modest demand: not a refusal of new technology, but a call for care and consultation before it is introduced. It is the kind of demand bargaining is designed to handle.
Evidence two: public employers are structurally open to negotiation
Public-sector workers are in a stronger position than many private-sector workers to shape AI adoption, for structural reasons.
First, government employees in many US states and cities are covered by collective bargaining rules that require employers to negotiate over changes to working conditions. New AI systems that monitor staff, allocate work or feed into evaluations can reasonably be argued to fall within that scope. Where this applies, a city cannot simply switch on a new tool without at least the possibility of a challenge.
Second, governments face political accountability that private firms do not. City councils and mayors answer to voters, and a strike by municipal staff disrupts public services directly. This raises the cost of introducing AI in a way workers regard as unfair, and gives unions leverage in negotiations.
Third, public bodies tend to adopt technology through formal procurement, which leaves a paper trail and creates points where unions can intervene. Contracts with AI vendors, pilot schemes and policy documents can be requested, scrutinised and debated. In a private company, comparable decisions may be made internally with little outside visibility.
The Guardian’s report frames this as a wider pattern of workers gaining ground as governments adopt AI. Beyond the Oakland case, the source material does not say which other jurisdictions or unions are involved, or what agreements have been reached, so the full extent of that trend cannot be confirmed here.
Evidence three: contract language is suited to the harms workers fear
The specific harms raised in Oakland, namely surveillance, inaccurate evaluation, and dismissal or discipline driven by AI, are well suited to contractual solutions.
A clause stating that no employee may be disciplined or dismissed solely on the basis of an automated system’s output is simple to write and simple to enforce. If the employer breaches it, the worker can file a grievance. Similar clauses can require advance notice before new monitoring tools are introduced, give workers access to the data held about them, or require that any automated performance assessment be reviewed by a person.
This kind of protection already has precedents outside the public sector. Unions representing screenwriters and actors in the US entertainment industry negotiated terms covering the use of AI after industrial action, which established that AI could be treated as a bargaining subject in its own right. Those agreements showed that unions can set limits on how employers use AI without blocking the technology entirely.
Legislation, by contrast, has to be general enough to cover many industries and employers, which often makes it less precise. It can also take years to pass and be applied. A contract can be renegotiated every few years as the technology changes, which makes it a flexible tool for a fast-moving field.
The strongest case against: bargaining protects few workers, and wins are unproven
The case against this argument is substantial and deserves a fair hearing.
First, a strike authorisation is not a victory. The Oakland vote shows that workers are willing to act, but the source material does not report that any AI protections have been agreed. Calling this a case of workers “winning” is premature until contract terms are settled and published. Employers may concede on pay and other issues while offering only vague commitments on AI, or none at all.
Second, collective bargaining reaches only workers who belong to unions. In the United States, most private-sector employees are not covered by a collective agreement. Gig workers, contractors and staff at non-union firms, who may be most exposed to algorithmic management, gain nothing directly from public-sector contracts. If protection depends on union membership, it will be uneven, and the workers with the least bargaining power will receive the least.
Third, contract clauses can be weak in practice. Saying that a human must review an AI-generated decision means little if the reviewer approves the system’s recommendation as a matter of routine. Enforcement relies on workers knowing when AI has been used, which may not be obvious. Without technical expertise and access to vendor systems, unions may struggle to check compliance.
Fourth, there is a public-interest argument on the other side. Governments are under pressure to deliver services efficiently, and some AI tools could reduce backlogs or improve accuracy. Restrictive bargaining terms could slow useful adoption, with costs borne by residents rather than by city management. Some would argue that decisions about public technology should rest with elected officials and the wider public, not mainly with the workforce.
Finally, legislation may in the end prove more effective. A well-designed law applies to everyone, union member or not, and can set baseline standards for transparency, data protection and human oversight. On this view, bargaining is a useful stopgap, but it cannot replace regulation.
Evidence that would change the conclusion
Several developments would strengthen or weaken the argument that collective bargaining is workers’ main defence against AI.
The argument would be strengthened if published agreements in Oakland or other cities contain specific, enforceable AI clauses, and if grievances brought under those clauses succeed. Evidence that such terms spread across many public employers, rather than staying in a handful of cases, would also support it.
It would be weakened if negotiations end with only general statements of intent on AI, or if AI-related demands are dropped in exchange for other concessions. Evidence that contractual protections are routinely bypassed, for instance through outsourcing work to vendors or contractors, would also undermine it.
The conclusion would also shift if comprehensive legislation on AI in employment were passed and enforced at state or federal level. In that case, the law rather than the contract might become the main source of protection, and bargaining would play a supporting role.
Independent research would help as well. Studies comparing unionised and non-unionised workplaces, looking at how AI is introduced, how often it informs disciplinary decisions and how workers fare, would give a clearer picture than individual disputes. That kind of evidence is not yet available in the material reviewed here.
Sources and further reading
- The Guardian, technology section: reporting on US workers and unions seeking AI safeguards as governments adopt the technology, including the Oakland strike authorisation.
- City of Oakland public records: council materials and policy documents on technology procurement and labour relations.
- US labour relations research institutes: academic work on collective bargaining in the public sector and on algorithmic management.
- Entertainment industry union agreements: published summaries of contract terms covering the use of artificial intelligence.
Surfaced from the rss:guardian_tech signal “unions bargaining over workplace AI”. AI-assisted draft, editorially reviewed.
