Career & Jobs
How Do You Handle a Colleague Who Quietly Relies on AI for Their Work?
To handle a colleague who quietly relies on AI for their work, separate the tool from the deliverable. Focus your feedback entirely on objective quality, accuracy, and operational impact rather than accusations of cheating. Initiate a direct, private conversation about specific errors, establish mutual verification standards for team handoffs, and agree on clear quality checkpoints before escalating anything to management.
Unannounced AI usage is now common across office environments. Some team members use language models to draft client communications, write internal briefs, or assemble data analyses in minutes. The tension arises when unedited synthetic text creates silent errors, breaks formatting standards, or forces peers to spend hours quietly correcting flawed output.
By Jim Vernon, Editor, AI Intelligence International · Published 8 September 2026 · Reviewed against our editorial standards · About the author
What are the key takeaways?
- Address the factual quality and accuracy of the output rather than policing the tools used to produce it.
- Document the measurable human time required to correct sloppy automated drafts before raising structural issues.
- Establish clear team verification standards so nobody can hand off unchecked raw model generations as finished work.
- Distinguish between helpful productivity acceleration and careless delegation that transfers work onto colleagues.
What does this article cover?
| Question answered | How Do You Handle a Colleague Who Quietly Relies on AI for Their Work? |
|---|---|
| Topic | Career & Jobs |
| Reading time | About 7 minutes (1,477 words) |
| Written by | Jim Vernon, Editor, AI Intelligence International |
| Published | 8 September 2026 |
| Last updated | 8 September 2026 |
Why does hidden AI use cause friction in teams?
Friction rarely starts because an employee uses an artificial intelligence model to speed up their research or initial drafting. It starts when that employee stops reading the generated text and hands it down the line as if it were finished work. Colleagues downriver suddenly encounter confident hallucinations, hollow corporate clichés, and subtle factual errors that demand forensic proofreading.
The emotional strain comes from an unfair transfer of cognitive labour. When one person generates twelve pages of project specifications in three minutes, they look productive on surface-level velocity metrics. However, if two colleagues must spend five hours auditing the text to ensure client safety, the organisation has not become faster. It has simply outsourced editorial drudgery to peers without their consent.
Secrecy also erodes operational trust. If team members suspect that a colleague is silently pasting internal company data into external consumer models, governance concerns arise. People wonder whether proprietary spreadsheets, customer records, or confidential roadmaps are leaking into public model training runs without managerial approval.
How do you distinguish between good AI use and reckless shortcuts?
Good artificial intelligence use is invisible in the finished product because human editorial judgment remains fully engaged throughout the process. A competent professional uses models to overcome blank-page inertia, synthesise raw interview notes, or suggest structural outlines. They cross-examine every assertion, verify citations against primary records, and adapt tone to match company conventions.
Reckless usage leaves obvious fingerprints across every deliverable. You will notice unnatural balance words such as 'delve', 'testament', or 'beacon', alongside repetitive paragraph structures that sound plausible while saying very little. More seriously, you will discover non-existent regulatory clauses, inventively merged numerical figures, and fabricated customer quotes that collapse under five seconds of scrutiny.
The dividing line is ownership. A productive operator treats a model as a fast, erratic intern whose drafts must be checked line by line. A reckless operator treats the model as an infallible replacement for their own critical reasoning, hitting copy, paste, and send without taking personal responsibility for the outcome.
What does this problem actually cost in wasted team hours?
The cost of careless automation is measurable in salary expense and diverted focus. Consider an operations team of four people, where an analyst named Mark produces weekly customer trend reports. Mark previously spent 8 hours researching and drafting an 8-page report, which his senior peer Sarah could review and sign off in 1 hour. The total weekly time invested was 9 hours.
Mark begins generating the report with an untuned model in 30 minutes, cutting his personal effort by 7.5 hours. However, the generated report now contains subtle metric inconsistencies, unverified product codes, and repetitive filler. Sarah can no longer skim the document; she must cross-check every single data table against the internal database.
Sarah's review time expands from 1 hour to 4.5 hours each week. In this scenario, Mark saved 7.5 hours, but Sarah lost 3.5 hours of high-value strategic time doing forensic cleanup. Across a 48-week working year, Sarah loses 168 hours to manual verification work that Mark skipped. If Sarah's hourly cost to the company is £45, that sloppy delegation wastes £7,560 of senior capacity annually on a single recurring document.
How should you start a direct conversation with your colleague?
Keep the initial discussion strictly private, informal, and anchored in concrete examples rather than speculation. Never open with an accusatory question like 'Did you write this with ChatGPT?' Such questions trigger immediate defensiveness, prompting the person to lie or shut down. Focus entirely on the tangible defects in the shared deliverable.
Present specific passages and explain the operational impact calmly. For example, say: 'In section three of the technical brief, the system architecture references an API endpoint that our platform decommissioned two years ago. I spent two hours trying to reconcile that diagram before realising it was invalid. What was the source for that architecture table?'
This framing forces the conversation onto facts. If they acknowledge using a generative model, respond constructively: 'The speed is helpful, but unverified drafts cannot enter the team backlog. If you use a tool to assemble the first pass, we need an agreed check to ensure every single link, number, and dependency is confirmed before you tag me for review.'
How can teams build clear boundaries without banning helpful tools?
Banning generative software outright never works; it simply drives adoption further underground. A healthy team creates explicit, written ground rules for AI-assisted deliverables. The foundation of any practical policy is simple: the human whose name is on the document remains 100% accountable for every word, figure, and claim inside it.
Establish explicit definitions of what constitutes a review-ready handoff. For instance, your team can agree that no document with unverified source links may be passed to another colleague. You can mandate that raw model outputs belong exclusively in personal scratchpads, never in shared workspace folders or pull requests.
Encourage transparency by removing the stigma around automation. When high performers openly discuss which prompts they use to draft customer update emails or summarise spreadsheets, others realise they do not need to hide their workflows. Openness allows the entire group to critique prompts, spot edge-case failures, and collectively raise the floor of quality.
When is it necessary to escalate the issue to your manager?
Escalation becomes mandatory when direct peer feedback fails, when client trust is threatened, or when regulatory compliance is breached. If you have already pointed out specific errors privately and the colleague continues to submit unchecked, hallucinatory drafts that derail sprint goals, you are no longer dealing with a learning curve. You are dealing with neglected duties.
When speaking with your team lead, present clear evidence of the operational bottleneck. Do not frame the issue as a moral dispute about machine assistance. Bring three specific documents, highlight the uncorrected errors, and outline the exact hours your team spent remediating the work before it could reach production.
Show your manager how the bottleneck affects team deliverables and deadlines. You might say: 'We want to support faster drafting methods, but unchecked automated output has added twelve hours of rework to our delivery cycle over the last fortnight. We need clear team-wide quality gates so that incomplete drafts do not progress to final sign-off.'
What do people ask most about this?
Can you prove whether text was written by an artificial intelligence tool?
You cannot reliably prove synthetic authorship using automated AI detection tools alone. Commercial detectors frequently deliver false positives on formal human writing and false negatives on lightly edited model outputs. Instead of attempting to prove machine generation, focus entirely on provable human errors. Document factual inaccuracies, fabricated citations, circular reasoning, and discrepancies with internal records. An employee is accountable for delivering accurate work regardless of the software they used.
What should I do if my colleague's AI habit puts sensitive company data at risk?
Data protection and intellectual property risks require immediate, objective intervention. If you observe a peer pasting confidential client lists, non-public financial records, or proprietary source code into personal consumer AI interfaces, treat it as an active security incident. Point out the data security breach calmly and reference company data governance rules. If they do not immediately stop, alert your data protection officer or direct manager to prevent further leakage.
How can I avoid sounding like I am against productivity tools when I speak to my boss?
Frame your feedback around quality control, risk management, and overall team throughput rather than the adoption of technology itself. Explain that you welcome tools that accelerate initial thinking, but emphasize that the current handoff process transfers a heavy proofreading burden onto others. Emphasise that standardising human verification will allow the team to reap the real speed benefits of automation without sacrificing professional standards.
What if my manager is the one producing low-quality AI work?
Managing upward requires careful tact and a strong focus on shared business goals. When a manager distributes superficial or contradictory model-generated briefs, ask clarifying questions that expose the gaps without direct confrontation. Ask: 'To ensure we deliver this accurately, could you clarify how this proposed timeline aligns with the third-quarter resource caps?' Detailed operational questions naturally require human reasoning, leading managers to reconsider raw model outputs.
How was this article researched?
This article is written and maintained by Jim Vernon, Editor at AI Intelligence International. Figures and claims are drawn from the calculators and models published on this site, from vendor documentation current at the time of writing, and from first-hand testing of the tools described. Every article is reviewed against our editorial standards before publication and re-checked whenever the underlying tools or pricing change.