Cannabis Workers Just Won AI Protections in Their Union Contract. Here’s What It Means for Every Cannabis Employer.

In late August, cultivation and processing workers at New England Treatment Access (NETA) in Franklin, Massachusetts ratified a new one-year contract with UFCW Local 1445. The agreement covers eighteen workers in cultivation, harvest, cure, and pest management roles. It raises wages across the board, adds a one-time $700 ratification bonus, and improves holiday pay, overtime pay, and benefits. It also adds successorship language that protects workers if new ownership buys the company. And it includes something most cannabis operators haven’t had to think about yet: a clause addressing AI in the cannabis workplace.

That clause is the real story here. Not because it’s dramatic on its own, but because it’s the first cannabis employer we’ve seen agree to it in writing. This is a good moment to separate three questions that tend to get blurred together. What does NETA’s contract specifically require? What does federal labor law require more broadly for any unionized workforce? And what applies to you if yours isn’t unionized? Those are three different answers, not one. Getting AI in the cannabis workplace right starts with knowing which one you’re actually answering.

What Actually Happened in Franklin, Massachusetts

The negotiation took 16 months to reach a deal. It had stalled before workers moved it forward with public pressure: wearing union apparel on shift and handbilling outside the facility, including during severe weather. According to UFCW’s own announcement of the ratified agreement, the final contract includes two provisions related to technology. In the union’s own words, it adds language “to prevent employees from being negatively impacted by an automated or artificial intelligence system.” It also requires the company to give the union advance notice of significant technology updates.

Worth being precise about what we know and don’t know here. UFCW’s public announcement describes the clause; it doesn’t publish the contract’s exact legal text. We haven’t seen the underlying agreement. So this article sticks to the union’s own description of what the clause does. It doesn’t assert more specific language, like a guaranteed protection of an employee’s “job, hours, or standing,” that we haven’t independently confirmed. If the precise contract language matters for your situation, talk to labor counsel with access to the actual document. A blog post can’t answer that for you.

What NETA’s Agreement Specifically Requires, and What Federal Law Requires More Generally

These are not the same thing, and conflating them is the easiest way to over-promise or under-prepare.

What NETA’s Contract Specifically Requires

NETA’s contract requires two specific things. First: advance notice to UFCW Local 1445 before the company rolls out significant technology updates. Second: the general protection UFCW describes against automated or AI systems that negatively impact employees. That obligation exists because NETA’s own collective bargaining agreement spells it out. It doesn’t automatically exist at any other cannabis operator, union or not, unless that employer’s own contract says so.

What Federal Law Requires for AI in the Cannabis Workplace Generally

Federal labor law requires something broader and less specific for any unionized employer with AI in the cannabis workplace. Under the National Labor Relations Act, an employer with a recognized union has to bargain in good faith. That duty covers “wages, hours, and other terms and conditions of employment”: what the NLRB calls mandatory subjects of bargaining. A unilateral change to a mandatory subject, made without bargaining first, can violate the Act on its own. The Board has treated some technology changes, including monitoring and surveillance technology, as falling within that duty to bargain. In December 2024, the Board’s Endurance Environmental Solutions decision restored a stricter “clear and unmistakable waiver” standard for these situations. That decision reportedly involved employer-installed vehicle cameras. It overruled a 2019 precedent that had let employers point to a generic management-rights clause as cover for skipping bargaining. Under the restored standard, a union has to explicitly and clearly waive its right to bargain over a specific kind of change. Only then can an employer rely on a contract clause to make that change unilaterally.

Here’s the part that matters for your specific situation. Whether a given AI or automation rollout actually triggers a bargaining obligation depends on your own CBA’s language and the specific facts of the change. So does what “advance notice” or “bargaining” looks like in practice. It’s not a blanket rule that every unionized employer must give notice before introducing any AI tool, the way NETA’s contract specifically requires. If your workforce is unionized and you’re evaluating AI, scheduling software, monitoring tools, or automated systems, read your own CBA’s management-rights and technology-change language with labor counsel first. Don’t assume you’re covered, and don’t assume you’re exposed. Check.

Where AI Is Already Showing Up in Cannabis Workplaces

Before getting into risk, it’s worth being concrete about where this actually shows up on the ground. AI in the cannabis workplace covers a wider range of tools than most operators picture. It’s also worth distinguishing AI, specifically, from ordinary automation. A scale that weighs product, or a timer on a grow light, isn’t an AI system. Software that scores, ranks, predicts, or makes a recommendation about a person, based on patterns in data, generally is.

We’re seeing cannabis operators use, evaluate, or get pitched on tools for:

• Resume and applicant screening
• Labor forecasting and shift scheduling
• POS-driven sales and performance analytics
• Employee productivity tracking
• Video surveillance and loss-prevention analytics
• Cultivation automation and environmental controls
• Performance review drafting
• Corrective-action and discipline documentation
• Workforce reduction recommendations
• Employee-facing HR chatbots

Some of these are genuinely AI: a system that scores which applicants look most like your best past hires. Some are automation with an AI layer bolted on for convenience, like scheduling software that now includes a “recommended shift” feature. The HR exposure shows up wherever the tool’s output touches a decision about a real person, not wherever the word “AI” appears in the vendor’s marketing.

Before You Buy the Tool

Buying the software is the easy part. The HR question starts when that software begins influencing someone’s schedule, performance rating, pay, or job.

The Pre-Purchase Checklist

Before signing a vendor contract for any tool that touches employment decisions, work through this list:

• What employment decision does the tool actually influence: hiring, scheduling, pay, discipline, promotion, or something else?
• What employee or applicant data does it use to make that recommendation?
• Where is that data stored, and who else can access it?
• Does employee or applicant data get used to train the vendor’s model, and can you opt out?
• Can the vendor explain, in plain language, why the tool produced a specific result for a specific person? If they can’t, that’s worth knowing before you sign, not after a candidate or employee asks.

The Rest of the Checklist: Accountability and Notice

• Can a human actually override the tool’s recommendation, and does anyone actually do that in practice, or does the recommendation just get rubber-stamped?
• Has the tool been tested for bias or disparate impact across protected groups, and will the vendor show you that testing?
• How does the tool handle disability accommodation requests, and does it flag them for human review rather than scoring around them?
• What notice, if any, does an employee or applicant get that the tool is being used on them?
• If your workforce is unionized, does this rollout trigger notice or bargaining obligations under your CBA?
• Does the tool trigger any state or local AI employment law? New York City’s Local Law 144 requires an independent bias audit and candidate notice before certain automated hiring tools are used on NYC applicants.
• Illinois has separate notice and consent requirements for AI-analyzed video interviews. Colorado’s AI Act, in effect since June 2026, requires impact assessments and consumer notice for “consequential” AI-driven employment decisions. If you operate in more than one state, this is a growing patchwork, not a settled one. Check state by state; our Multi-State Cannabis HR Playbook covers how we approach exactly this kind of compliance sprawl.
• If the tool gets something wrong, whose job is it to catch it, and who’s actually accountable, the vendor or your own HR team? “The vendor said it was fine” is not a defense on its own.

If the vendor cannot explain what employee data the system uses, that is a question to answer before signing the contract, not after rollout.

AI Can Draft. It Should Not Decide.

A lot of the real, day-to-day AI in the cannabis workplace isn’t a fancy screening algorithm. It’s a manager typing a rough set of facts into ChatGPT, Claude, Copilot, or an HR platform’s built-in assistant. Then asking it to write a corrective action, a performance review, an investigation summary, or a termination letter.

That’s not inherently a problem. These tools are genuinely useful for turning messy notes into something clear and professionally worded. The problem shows up when the tool’s polish gets mistaken for judgment. AI can draft a corrective action. It cannot tell you whether the facts your manager gave it are accurate. A manager’s incomplete or mistaken account of what happened still produces a bad corrective action, even when AI writes it well. It’s now just a bad corrective action with a more convincing paper trail behind it. The employment decision, and the responsibility for getting the facts right, still belongs to the manager and to HR. The tool is a drafting assistant, not a witness.

Human Review Needs to Be Real

This matters enough to say plainly: a manager clicking “approve” on an AI-generated recommendation, without actually engaging with it, is not meaningful human review. It’s a rubber stamp with a human fingerprint on it, and if that recommendation is ever challenged, “the software suggested it” won’t hold up as an explanation.

Real review means the person approving a decision actually did the work. And they asked whether the recommendation actually makes sense given everything they know, not just whether it looks plausible. If your reviewers can’t describe what they checked, they didn’t check it.

Automation Can Change Jobs Even Without Generative AI

This part matters especially for cultivation and manufacturing. The technology conversation tends to focus on chatbots, and it completely misses where automation is actually reshaping jobs. Automated irrigation and environmental controls, trimming and processing equipment, inventory and seed-to-sale systems, and scheduling technology are all changing what cultivation and production roles require day to day. None of that needs to involve anything you’d call “AI.”

None of it needs a language model to raise the same HR questions. When a piece of equipment or a new system changes what a role requires, it can affect duties and staffing levels. It can affect hours, the skills the job actually needs, and how performance gets measured. Say a cultivation tech’s job used to be mostly manual monitoring, and it’s now mostly overseeing an automated system. That’s a real change to the job. Document it and communicate it deliberately. Don’t let the job description quietly drift out of date.

AI in the Cannabis Workplace: What It Means If You’re Unionized

If your workforce is unionized, the practical takeaway isn’t “you now owe advance notice before any AI rollout.” That specific obligation belongs to NETA’s contract, not to a universal rule. What is universal is the underlying NLRA duty to bargain in good faith over changes to mandatory subjects: wages, hours, and working conditions. The Board has found that some technology changes, including monitoring and AI-assisted tools, fall within that duty. Before you introduce AI scheduling software, automated cultivation systems, or AI-assisted discipline decisions, read your own CBA’s technology language with labor counsel. Read the management-rights clause too. Loop counsel in before you sign a vendor contract, not after.

AI in the Cannabis Workplace: What It Means If You’re Not Unionized

Non-union employers don’t have a CBA to comply with, so there’s no contract clause requiring advance notice the way NETA’s does. But that doesn’t mean the exposure is the same exposure in a smaller form. It’s a genuinely different kind of exposure, rooted in anti-discrimination law rather than labor law.

Federal anti-discrimination law, chiefly Title VII, the ADA, and the ADEA, applies to AI-driven employment decisions the same way it applies to any other employment decision. That covers hiring and applicant screening, employee monitoring, productivity measurement, scheduling, pay, promotion, discipline, layoffs, termination, and disability accommodation. A tool doesn’t have to be designed to discriminate to create legal exposure. That’s the disparate impact theory: a facially neutral tool can still create liability if it produces a statistically skewed outcome for a protected group. Protected grounds include race, color, religion, sex (including pregnancy, sexual orientation, and gender identity), national origin, age 40 and over, disability, and genetic information. Intent doesn’t matter here. And using a third-party vendor’s tool doesn’t transfer that risk away from you. A vendor’s assurance that its tool is compliant doesn’t shield the employer that actually uses it to make employment decisions.

The Withdrawn EEOC Guidance, Explained

One nuance worth being straight about: the EEOC withdrew its own technical-assistance documents on AI. That included a 2022 document on AI and the ADA, a 2023 document on AI and Title VII, and a December 2024 fact sheet on workplace monitoring technology. All three came down from its website in early 2025, following a change in administration and a new federal AI policy direction. That doesn’t mean the underlying laws changed. Title VII, the ADA, and the ADEA are statutes, not guidance documents. They still apply in full to AI-driven employment decisions, regardless of whether the EEOC currently hosts a standalone AI page. What’s shifted is that federal enforcement priorities and published guidance are less settled now than they were in 2023. State and local governments have been filling part of that gap with their own AI employment laws, the kind referenced in the checklist above.

The practical version, union or not: build the same discipline voluntarily that a union contract would otherwise force. Document what tools you use. Test for disparate impact before and after rollout. Keep a human genuinely in the loop, and don’t wait for an employee to raise a concern before you’ve thought through the answer. If wage and hour exposure is also on your radar this year, our Q3 compliance audit guide walks through the broader compliance picture AI tools now sit inside.

What Should an AI Workplace Policy Cover?

A real AI workplace policy needs to be more than a page in the handbook that says “be responsible with AI.” At minimum, it should address:

• Which AI tools are approved for use, and by whom
• Prohibited uses, including feeding confidential business or client information into public AI tools
• How applicant and employee data can and can’t be used, including whether it can train a vendor’s model
• Which employment decisions an AI tool can inform versus which ones require a human decision-maker
• What human review actually has to look like before an AI-influenced decision is finalized
• How new AI tools get vetted and approved before a manager or department starts using one on their own
• What notice employees or applicants receive when an AI tool is part of a decision that affects them
• How monitoring and surveillance tools are disclosed and limited
• How the policy handles accommodation requests that intersect with an AI-driven process
• Where employees or applicants can raise questions or complaints about an AI-influenced decision
• What gets documented, and for how long
• What managers are responsible for versus what HR is responsible for

Getting AI in the cannabis workplace right starts with writing this down before you need it, not after.

Frequently Asked Questions

What AI protections are in the NETA union contract?

Per UFCW Local 1445’s own announcement, the contract includes technology provisions that prevent employees from being negatively impacted by an automated or artificial intelligence system. It also requires NETA to give the union advance notice before rolling out significant technology updates. The union’s announcement describes the clause; nobody has independently reviewed the full contract text for this article.

Can employers use AI to screen job applicants?

Generally yes, but the same anti-discrimination laws that apply to any screening process apply to an AI-driven one. If the tool produces a disparate impact on a protected group, that can create liability under Title VII or the ADA regardless of intent. Some states and cities, including New York City, have added specific notice and bias-audit requirements on top of that federal baseline.

Can employers use AI for employee discipline?

AI tools can help draft or organize discipline documentation, but the decision itself, and responsibility for the underlying facts, should stay with a human manager and HR. If your workforce is unionized, a shift toward AI-assisted discipline may also raise bargaining obligations depending on your CBA and whether the change affects a mandatory subject.

Can employers use AI for performance reviews?

Similarly yes, with the same caveat. AI can help draft language, but it shouldn’t be the source of the underlying judgment about someone’s performance. A manager approving an AI-drafted review needs to actually verify what it says against what they know.

Should employees be told when AI is being used?

It depends on your state, your industry, and what the tool does. But as a baseline practice, yes. Several state and local laws already require notice for specific uses, like automated hiring tools or AI-analyzed video interviews. Voluntary transparency also tends to prevent the kind of blindsided reaction that turns into a complaint or a grievance.

Can employers use AI to monitor employees?

Yes, but the NLRB has specifically treated monitoring and surveillance technology as something that can fall within a unionized employer’s duty to bargain, particularly where it could be used for discipline. Non-union employers face separate exposure if monitoring data feeds into decisions that end up having a disparate impact on a protected group.

What should be included in an AI workplace policy?

At minimum: approved and prohibited tools and uses, how applicant and employee data can be used, and which decisions require human judgment. Add what real human review looks like, a vendor approval process, employee notice, and monitoring disclosure. Round it out with accommodation handling, a complaint channel, documentation requirements, and clear manager versus HR responsibilities. See the full framework above for building AI in the cannabis workplace responsibly.

Do unionized employers have to bargain over AI or automation?

Not automatically, and not as a blanket rule. Under the NLRA, an employer with a recognized union has to bargain in good faith over changes to mandatory subjects: wages, hours, and working conditions. Unilateral changes to those can violate the Act. The NLRB has found in specific cases that technology and monitoring changes fall within that duty. Whether a specific AI or automation rollout actually triggers bargaining depends on your own CBA’s language and the facts of the change. NETA’s contract now includes a specific advance-notice requirement. That’s a contractual term of that agreement, not a rule that automatically applies everywhere.

Want a second set of eyes on the policy side before you commit? Book a 15-minute call. Building out an employee handbook that actually addresses AI in the cannabis workplace? Our Cannabis Employee Handbook Guide covers what belongs in it. And our Cannabis HR Services page covers what fractional support looks like if you’d rather have someone else build the policy with you.