Meta recently launched an AI agent called Muse that can send emails, book holidays and fill out online forms, and it is drawing attention for a capability no major consumer agent has offered at scale: hunting for deals and completing purchases without explicit user sign-off S¹. Google's shopping blog confirms a parallel push, with VP Vidhya Srinivasan announcing a Universal Cart and what she calls "the foundation for agentic commerce" P⁵ — a vendor's own announcement with no independent testing cited. OpenAI is developing shopping tools as well, according to the same reporting S¹. None of these capability claims have been independently verified by this publication.
None of these companies has published data on how often their agents select ethically sourced products, or whether they can. An AI assistant can compare prices, delivery times and customer reviews in seconds. Establishing whether workers are paid fairly, can join unions or have safe workplaces is a far harder problem, and no agent developer claims to have solved it S¹.
My read: This is the first consumer AI wave I've seen where the gap between what the technology optimises and what it cannot measure is the entire story. I don't buy that any of these companies have solved the ethics-verification problem, because the evidence shows even human auditors struggle to see past subcontracting layers. What I would watch is whether any agent developer builds a labour-conditions data layer before these tools reach scale, or whether bargain-hunting speed becomes the only metric that matters.
Buyers' price pressure already squeezes workers
The bargain-hunting these agents automate is not new.
It is just faster, and applied at a scale no human shopper can match. Research by the International Labour Organisation, covering 31 suppliers across five countries, found that buyers' demands already put pressure on wages, increased overtime and sometimes led factories to cut corners on workplace safety S¹. Those findings are second-hand in this source, without a direct citation to the original ILO publication, so the specifics warrant checking against ILO records. Australian labour policy expert David Peetz has explored how interconnected technologies can concentrate economic power S¹, and UK researchers have demonstrated how consumer relationships and global supply chains shape the governance of labour standards S¹.
The common thread: when the buyer has more power and less visibility, the worker at the end of the chain absorbs the cost.
Companies' ethical sourcing claims may reveal little about what happens in subcontracted factories S¹. A premium price on running shoes often reflects branding or profit margins rather than better wages or conditions S¹. Worldwide, tens of millions of people live in modern slavery, though the figure is vague and unattributed to a specific report in the source S¹.
Shopping agents inherit human blind spots
Researchers at McGill University, Davood Wadi and Yu Ma, are studying how agentic AI deploys human heuristics as a surrogate consumer P². Their work suggests shopping agents may replicate human decision-making shortcuts, including the blind spots that lead shoppers to optimise for price and speed over provenance. No error bars or released code accompany this research in the cited source.
This matters because the agents are being positioned as autonomous actors, not comparison tools. Meta's Muse has the potential to make purchases without approval, though it has not yet been observed doing so S¹. Google's Universal Cart, announced on the company's official blog, is explicitly framed as the first step toward agents that complete transactions P⁵.
A procurement team testing an agent today could verify price and delivery accuracy against a supplier catalogue in minutes. Verifying that the same supplier's factory pays a living wage would require audit data the agent cannot fetch from a product page. For a category manager at a mid-size retailer, the practical shift is that an agent can now surface the cheapest supplier across dozens of markets in seconds, but the same manager still has no automated way to check whether that supplier meets the company's labour code — that verification still requires a human audit cycle measured in weeks.
Faster purchasing, unchanged visibility
No empirical evidence yet directly links AI shopping agents to increased labour exploitation S¹. The risk is structural: agents that automate price pressure at greater speed and scale, with no corresponding mechanism to verify labour conditions, could deepen a dynamic the ILO has already documented in human-driven purchasing S¹.
Google's Universal Cart is live on the company's shopping platform P⁵, Meta's Muse has launched S¹, and OpenAI's shopping tools remain in development S¹. The next checkpoint is whether any of these platforms publishes a method for filtering or ranking products by labour standards before agents begin transacting at scale.
Sources: S1 — New AI shopping agents promise better bargains. But who pays the price · P2 — Shopping by algorithm: How agentic AI deploys human heuristics as a su · P3 — International Labour Organization · P4 — peetzweg/ · P5 — Google Shopping introduces Universal Cart, agentic shopping
Written from 5 sourced items, 4 of them primary.
