Category: Governance

How AI is being used to make decisions about people in the UK: housing, hiring, education, policing, finance. We track power, accountability, and harm, in public.

  • Who Owns Your Work Data When AI Is Watching

    Who Owns Your Work Data When AI Is Watching

    Your company says AI will “boost productivity”; however, it also quietly captures keystroke data, tool usage, message tone, and output speed.

    That data does not float in the air. Someone owns it. Someone queries it. Someone can profile you with it.

    What is “work data” today

    • Activity traces: Logins, app usage, files opened, meeting joins
    • Communication signals: Email or chat metadata, sentiment, response times
    • Output artefacts: Drafts, code diffs, slide versions, tickets
    • Derived scores: Productivity, risk, “influence”, compliance

    Default reality

    • The company owns work systems and their telemetry.
    • Vendors process that data under contract.
    • You often do not see derived scores that affect opportunities or reviews.

    Why is this risky for you

    • Models freeze early impressions. Yesterday’s sprint becomes today’s label.
    • Scores leak beyond their original use, such as promotions, assignments, and terminations.
    • Appeals are informal or nonexistent.

    Your five questions for HR and IT

    1. What exact data sources feed into any productivity or risk scoring? Name them.
    2. Who can see raw traces versus just aggregates? List job roles.
    3. What decisions may use these scores? Promotion, Performance Improvement Plan (PIP), Reduction in Force/Layoffs (RIF), pay, and access.
    4. How can I correct errors? What is the human review route and Service Level Agreement (SLA)?
    5. How long do you retain my traces and derived scores? Do you delete on request?

    Minimum viable policy you should ask for

    • Published inventory of monitoring tools and models.
    • Documented human review process with timelines.
    • Ban on using monitoring data for health or off-work activity.
    • Access log: Who viewed my data and when.
    • Retention limits: Raw traces short, derived scores audited.

    Manager checklist

    • Tell your team what is captured in writing.
    • Use monitoring to remove friction. Not to micromanage.
    • Review cases with two humans when scores drive outcomes.
    • Allow people to review and correct their records quarterly.
    • Reward documented impact. Not a noisy activity.

    You cannot manage trust with a black box.

    Get the weekly Human Defence Brief: plain English scripts to protect your job, your money, and your rights.

    Subscribe to our Newsletter.

    Latest posts

  • How to challenge an AI decision about you in the UK

    How to challenge an AI decision about you in the UK

    When to use this

    • You were denied a service or flagged at work, and the decision felt automated.
    • A payment, claim, loan, or moderation outcome arrived with no human you can reach.
    • You suspect monitoring or scoring is running in the background.

    The 90-second version

    1. Ask if automation was used: “Was an automated system involved in this decision, yes or no?”
    2. Ask for a human review: “I want a documented human review of my case.”
    3. Ask for an explanation: “Explain the main factors and data used. Include any accuracy score or confidence level if available.”
    4. Correct the record: “Here is new or corrected information that changes the outcome.”
    5. Get it in writing: Request a dated response from a named person, along with a contact route for appeal.

    Email template for first request

    Subject: Request for human review and explanation of automated decision

    Hello [Team or Contact],

    I received [decision or outcome] on [date]. Please confirm whether any automated system was used. If yes, I am requesting:

    • A documented human review of my case.
    • A plain-English explanation of the main factors and data used.
    • Any available accuracy or confidence information.
    • A route to appeal and the time limits.

    Here is the context that may change the outcome. [short bullet points only]

    Please reply in writing within [10 working days] with a named contact.

    Regards,
    [Name]
    [Email]
    [Reference number or account if relevant]

    Email template for escalation

    Subject: Escalation – No response yet to automated decision review

    Hello [Manager or Complaints Team],

    I requested a human review on [date] and have not received a sufficient response. Please escalate this as a formal complaint. I am again requesting:

    • A human review by a qualified person.
    • A plain-English explanation of the factors and data used.
    • A clear route to appeal with dates.

    If external escalation is needed, confirm the correct ombudsman or regulator.

    Regards,
    [Name]

    Short call script

    “Before we continue, please confirm whether automation was used. I am requesting a human review and a written explanation of the main factors. I will send this as an email now so we have a record.”

    What to attach

    • Screenshots or PDFs of the decision.
    • Dates and times of any calls or messages.
    • Only the facts that directly change the outcome.

    Red flags that justify escalation

    • No one will confirm whether automation was used.
    • You receive only a generic stock reply.
    • The explanation does not match the decision.
    • “Computer says no” with no appeal route or timeline.

    Manager’s view – If you run teams

    • Publish a human review route with a named role and a response SLA.
    • Keep an appeals log:- decisions, dates, outcome, fixes.
    • Never use “urgent” to override fairness or due process.

    Get the weekly Human Defence Brief: plain English scripts to protect your job, your money, and your rights.

    Subscribe to our Newsletter.

    Latest posts