The Co-operative Group — an organization built on mutual ownership and member democracy — is now using an OpenAI model to record, transcribe, and score every phone call made by its probate and wills advisers. The system evaluates more than 50 discrete aspects of each conversation, generating pass/fail scores that managers use to assess performance. Workers handling calls from recently bereaved customers report being monitored by AI for several hours a day. The irony of a co-operative deploying panoptic surveillance on its own workforce is apparently lost on management. This is not quality assurance in any traditional sense. Traditional call monitoring samples a handful of interactions per month for compliance purposes. What Co-op has built is total-coverage scoring: every word, every pause, every interaction fed through a model trained to judge whether the worker performed correctly across 50+ behavioral dimensions. The system is also understood to be used to identify ways staff can boost sales performance — meaning bereavement calls are being optimized for upselling, not just empathy. The pattern is accelerating. Euan Blair's Multiverse was revealed last week using AI for blanket monitoring of online classrooms. Burger King announced AI tracking of worker-customer interactions in February. Meta paused keystroke monitoring of employees this summer only after a staff backlash. Each deployment normalizes the next. The International Labour Organization warned this year about the psychosocial risks of AI workplace monitoring, including work-related stress, but noted the absence of any comprehensive regulatory framework to prevent harm. Co-op's defense is revealing in its hollowness. Managing director Caoilionn Hurley frames the system as enabling "empathetic expert guidance" and says "human judgment, accountability, care and kindness remain at the heart of every client relationship." But the system's architecture contradicts this: it replaces human judgment about call quality with algorithmic scoring, and it monitors every interaction rather than sampling. When a company says AI "supports" colleagues while scoring every word they speak, the language is doing extraction work — laundering surveillance as care. The trade union response highlights the structural problem. The Communication Workers Union's John Chadfield argues that "unaccountable computer systems should not be people's managers." But the deeper issue is that no UK law prevents this. The ILO framework is advisory. The EU AI Act classifies workplace monitoring as high-risk but does not ban it. Britain, post-Brexit, has no equivalent regulation in force. Employers are deploying these systems into a regulatory vacuum and daring workers to object. The Chartered Institute of Professional Development's framing is perhaps the most telling: "The office is becoming more like a factory now our performance feedback can be given in a more structured way." This is presented neutrally, but the history of factory-floor surveillance — from Taylorism to Amazon warehouse metrics — shows where structured performance feedback at total coverage leads: to workers optimizing for the metric rather than the outcome, to stress-related attrition, and to the hollowing out of the professional judgment the system claims to support. What makes this especially extractive is the asymmetry. Workers cannot opt out, cannot see the model's training data, cannot appeal a score, and cannot audit the 50+ criteria against which they're judged. Management captures the efficiency gains; workers bear the psychosocial costs. The bereaved customers calling about probate are unwitting participants in a system designed to score the person trying to help them. Value flows upward; stress flows down.
