Background and insight
MIT Technology Review describes a risk in organisations using multiple AI agents in work processes. One agent might retrieve customer pricing from several sources, including restricted systems, internal messages, and archived quotes. A second agent may then use the response to draft a proposal without anyone checking the handoff. The result can look credible even when its data is wrong, outdated, or not meant to be shared.
Data lineage means being able to trace where data originated, who changed it, which rules apply, and which systems used it. This makes it possible to investigate an error back through the chain and set clear limits on agents’ access. The technology does not replace human responsibility, but it can make oversight more systematic.
What does this mean for your everyday life?
For employees, AI agents can save time on routine tasks, but errors may also affect customer commitments, confidentiality, and trust. Clear checkpoints protect both workers and customers: users should know what information an agent relied on and who must approve decisions with consequences.
💡 How you can use this knowledge
- Before using AI for a work task, clarify which data sources it can access and whether customer data or internal documents may appear in its answer.
- Ask for citations or traceability for important figures, prices, and claims. Check them against the authoritative system before sharing anything.
- Require human approval for proposals, financial decisions, and sharing sensitive information; record who approved the result.