The Algorithm Doesn’t Get a Vote: Writing Your Co-op’s AI Rules in the Open

Someone will raise it at a members’ meeting. Maybe it’s the minute-taking, the newsletter, the grant application, or the chatbot a supplier keeps emailing about. Half the room wants a blanket ban. The other half has already quietly signed up to three tools and is getting quite attached to them. Both reactions are understandable. Neither is a policy.

A co-op AI policy shouldn’t be a scolding notice pinned to the staff room wall. It should be a shared agreement about how we use powerful tools without handing over the decision-making that makes us a co-op in the first place. Open books means being honest about what we use, why, and who benefits. Shared futures means not letting efficiency gains disappear into a vendor’s subscription line while members get the same workload and less say.

Start with what is already happening

Before drafting anything, do an honest inventory. Not a procurement exercise — a show-and-tell. Ask members and staff to list the tools they already use, even the boring ones: spellcheck, scheduling assistants, transcription, grant-finding tools, social media schedulers, the chatbot on the website. No shame, no gotchas. You cannot govern what you pretend isn’t there.

Separate the categories in plain English. There is a difference between automation (a rule that moves a form from inbox A to inbox B), predictive tools (software that scores or sorts things), and generative AI (tools that draft text, images, audio or code). The risks aren’t identical. A co-op that treats every tool as either ‘fine’ or ‘Skynet’ will write a policy nobody uses.

Where AI can help — with a human in the loop

Permitted uses should be specific enough to be useful. Routine drafting with a named human checker. Summarising public documents. Brainstorming options. Accessibility support such as transcription or translation, with a person reviewing the output. Research starting points that get verified. Scheduling and admin. None of these need to be dramatic.

Then set the second-pair-of-eyes rule. Anything member-facing, customer-facing, financial, legal, safeguarding-related or reputational gets human review before it goes out. Not a vague ‘sense check’ — a named role, a recorded sign-off, and a route to challenge. If a tool drafts a housing complaint response, a redundancy note, a grant bid or a safeguarding summary, a human being is accountable. The algorithm does not get a vote.

Prohibited uses should be equally clear. No automated final decisions on membership, discipline, credit, hiring, benefits, complaints or access to services. No publishing AI-generated content without checking facts, tone and rights. No pasting confidential information into a consumer tool. No using AI to surveil members or staff. No ‘shadow AI’ on personal accounts handling co-op data.

The data line: what never goes into a prompt

This is where UK co-ops need to be boringly strict. Personal data, special category data, member records, payroll, health information, safeguarding notes, complaint details, ID documents, unpublished finances and contracts should not be dropped into a public AI tool. ‘But it’s only a summary’ is how leaks happen. UK GDPR still applies. Data minimisation still applies. The ICO’s guidance on AI and data protection is not optional reading.

Use co-op-owned accounts with proper settings: training switched off where possible, retention limited, access controls in place. Do vendor checks — where data is processed, how long it is kept, whether subprocessors are involved, what happens on termination, and whether the contract gives you the assurances you need. If someone cannot explain where the data goes, the answer is no. If you are unsure, the answer is no. ‘It was public’ is not a licence to feed it into a model.

Disclosure without the confession booth

Disclosure should be plain, not performative. A line in an email footer — ‘Drafted with AI assistance, checked by the office team’ — is often enough. A short page on the website explaining where AI is used and where it isn’t. Minutes that record if a meeting was transcribed by a tool and who approved the final version. A note in a funding bid if AI helped draft it. Members should not have to play detective.

The test is meaningful use. If AI shaped a decision, a public message, a financial document or a member’s experience, say so. You do not need to disclose every autocomplete keystroke. You do need to disclose anything that affects money, people or power.

Productivity gains belong to the co-op

This is the bit most AI policies miss. If a tool saves five hours a week, where do those hours go? Into more output? Shorter hours? Better member support? Training? Or straight into an expectation that everyone does more with less? A co-op should decide that collectively, not let it happen by default.

Measure the right things: time saved, quality, stress, accessibility, member satisfaction, error rates. Then review the gains in the open. Put AI subscriptions and training in the budget. Pilot tools with clear stop-go criteria. Record what didn’t work. And watch for the quiet transfer of power — if only one person understands the tool, or the vendor controls the terms, the co-op has a problem.

Make the policy member-owned

Draft it with a small cross-section of members and staff, not just the tech-confident. Keep it to two pages plus an appendix. Use consent-based or consensus decision-making. Name a responsible person. Set a review date six months out. Report back to members on use, incidents and savings. Add a route for concerns that isn’t just ‘speak to the manager’. Require procurement checks before new tools are bought. Allow safe experimentation in a sandbox so people aren’t tempted to go underground. And make sure changes to the policy itself come back to members — not a quiet board edit.

A ban feels safer, but it usually just pushes use into private accounts and private chats. A free-for-all feels modern, but it erodes trust. An open, member-written policy is slower. It is also the only version that actually belongs to the co-op.

A starter rulebook, in plain English

• We use AI as an assistive tool, not an authority.
• A named human is accountable for every output that affects people, money or public claims.
• Member and customer data stays out of public AI tools unless a proper data protection assessment says otherwise.
• We disclose meaningful AI use in plain language.
• No automated decisions about membership, discipline, hiring, credit, safeguarding or access to services.
• Productivity gains are reviewed collectively and shared fairly.
• New tools go through a procurement and data check before use.
• Anyone can raise a concern without blame, and the policy is reviewed with members every six months.

That is not a ban. It is a boundary. And boundaries are easier to keep when everyone helped draw them.