How should employees be involved in decisions about AI?

Before the tool is chosen, with a real say in what it is for, and through people who are able to say no. That is the pattern in the one large agreement studied so far. It is also what the UK government consulted on until 30 September 2026. Kaiser Permanente and an alliance of unions representing 62,000 healthcare workers agreed in 2025 to put worker input into every stage of decisions about AI, with a joint task force of five from each side holding authority over technology investment. The UK consultation floated a legal requirement to consult and negotiate before monitoring technology is adopted, and the AI minister said in Liverpool that workers would have a voice. Most employers surveyed have not yet built an approval process for the tools, let alone for the people.

The agreement that put workers in the room#

Betsy Vereckey at MIT Sloan summarised on 28 September a report by Thomas Kochan, Erin Kelly and Arrow Minster on the 2025 negotiation between Kaiser Permanente and the Alliance of Health Care Unions. She calls the result “the first comprehensive agreement by a major U.S. healthcare employer and its unions to incorporate worker input into every stage of decision-making around how AI will be developed and used, from identifying problems to designing tools and training employees.” The machinery is described plainly. An AI task force “comprises 10 senior leadership members (five from labor and five from KP) who have direct decision-making authority over technology investments.” A network of “more than 3,500 unit-based teams” of frontline workers, managers and doctors looks for places where AI could help. Union representatives agreed to train peer advisers, colleagues who learn the tools first and then help others adapt. The order of events is the lesson the article draws: the parties succeeded by “starting with communication, then implementing the technology”, after a joint conference of about 300 union and 80 company leaders in February 2025. The report’s own verdict is that this “could not have been achieved through traditional position-based negotiations.”

What the UK consulted on#

The government’s consultation on workplace monitoring technologies, which closed on 30 September, makes worker engagement its third principle: “Meaningful worker engagement is central to the responsible and effective use of WMT.” Its example is modest and practical: “employers could pilot a new monitoring system with a group of workers or representatives and seek feedback before wider rollout, using that feedback to adjust thresholds, metrics or how outputs are interpreted.” Its second option goes further. “Primary legislation could be used to create a legal requirement that employers’ plans to adopt WMT would be subject to consultation and negotiation, with a view to agreement of trade unions or elected staff representatives where there is no trade union.” The process “would not require agreement to be reached in all cases”, and the document states no preference between that, a statutory code and plain guidance. It covers monitoring technology and not AI in general. The minister, Kanishka Narayan, was reported by the BBC, read in Finwire’s syndication, as “determined that workers would have a voice in the transition to AI”.

What the unions asked for in Liverpool#

The Labour conference’s list of contemporary motions carries one on AI, submitted by the Communication Workers Union. Two of its demands on the government concern decisions at work: “Give recognised trade unions the right to negotiate and agree the introduction and use of AI in the workplace.” And: “Ensure trade unions are embedded in decisions over the development, procurement and deployment of AI across public services.” The document does not record a vote and no result was found. Read beside the consultation, the motion marks out three different rights that are often spoken of as one. To be told is the first. To be consulted, with a view to agreement that need not be reached, is the government’s middle option. To agree, which is a right to refuse, is what the union asks for and what the Kaiser task force comes closest to, for investment decisions only. What unions elsewhere have said is collected on what trade unions and workers say about AI.

Where most employers are#

Three surveys reported this week suggest the starting point is lower than any of these. Stephen Owens at HCAMag reported: “Only 55% of employers have a formal review or approval process for AI tools, according to Littler Mendelson’s 2026 Annual Employer Survey of more than 300 U.S. executives, in-house lawyers and HR professionals.” Alexei Alexis at CFO Dive reported a vendor’s survey of 854 finance professionals in which fear of displacement ranked as the biggest barrier to adoption, and quoted the report: “They are hesitating because they understand it well enough to ask what happens next, and too few organizations have given them a credible answer.” Lara Ewen at HR Dive reported a consultancy’s survey of 200 US HR professionals in which 36 per cent said they were “often or always unclear” about how AI-generated results were achieved. All three are American and come from firms with services to sell: a law firm, a software vendor and a consultancy. They agree on one thing a manager can check locally: staff are being handed tools before anyone has told them what the tools are for.

Why involvement is a control#

The case for involving employees is usually made as fairness. The stronger case is that they hold information the organisation needs and cannot get another way. They know which parts of a task carry the judgement, which the estate maps on the tasks workers do not want automated. They know which routine work is where a newcomer learns, the subject of the missing rungs. They are the people expected to notice when the system is wrong, and the largest survey on whether employees know they are expected to override AI found that most do not count it as part of the job. And when they are left out they route around the policy, which the estate treats on working around the AI policy. An employer that decides alone is deciding with less information than it could have had, and is relying for oversight on people who had no part in the design.

Four things to settle with them before the next tool#

Rules Before Tools supplies the agenda, and the Kaiser sequence supplies the timing: the conversation comes first. Which decisions may the machine make in this team, settled with the people who make them now, and written down. Who can stop it: at least one person who does the work, with a route that does not run through the project’s sponsor. What people must remain able to do without it, which the team is best placed to name, since they know what they are ceasing to practise. And how anyone would know it had gone wrong: a channel for saying so that reaches someone with authority, and a reply. Each of the three rights in the Liverpool motion can be granted at a different point on that list. An employer need not wait for legislation to decide which it is offering, and saying which is more useful to staff than a promise of voice. The measurable risk sits in the handover of decisions to machines, and the people closest to the handover are the cheapest instrument for watching it.

What this does not show#

The Kaiser agreement was read through MIT Sloan’s summary; the report itself was not, and the summary gives no outcomes. Nothing shows the tools chosen were better, safer or more used than they would otherwise have been. It is one American healthcare employer with a long partnership with its unions, and the pattern may not travel. The UK document is a consultation with no decision, limited to monitoring technology. The motion’s fate at conference is unknown. The three surveys are American, each comes from a firm with an interest in the answer, and none gives its field dates in the reports read. No study has compared AI deployments made with and without employee involvement, which is the measurement the argument on this page needs and does not have.