AI at work you did not ask for

Workplace AI is the version of this module you have least control over. You did not choose the tools, you frequently cannot switch them off, and the data involved is not yours to decide about.

What you can do is know what is running and ask good questions.

Find out what is actually deployed

Most people work alongside AI tooling they could not name. Worth establishing:

Meeting tools — recording, transcription, summaries, and whether they run automatically.

Communication tools — summarisation, sentiment analysis, search across messages.

Productivity and performance tooling — activity metrics, “engagement” scores, code or output analysis.

Security tooling — content scanning, data-loss prevention, anomaly detection. Usually legitimate and usually broader than people assume.

Recruitment — screening for roles you might be hiring for. E8·L4’s accountability question lands directly here.

Ask IT, or your manager, or read the intranet. The answer is usually available and usually unread.

Policy or rumour?

The most common workplace situation is a confident belief with no document behind it — “apparently they can read everything”, or equally “nobody’s looking”. Both are guesses.

Ask for the written policy: what tools are in use, what they process, how long it is retained, who can see it, and what it may be used for. Most organisations of any size have one, often for compliance reasons, and often nobody has read it since induction.

If there is no policy, that is itself the finding, and worth saying to somebody who can commission one.

E4·L5 established that recording needs everyone to know and agree, and that it is a legal requirement in many places. Inside an organisation this does not dissolve — colleagues are people, and automatic transcription in a conferencing tool is still recording.

Practical position: say at the start that it is on. If a tool records by default, the announcement is not optional politeness, it is the thing that makes it lawful in many jurisdictions.

If someone objects, that is information rather than obstruction. There are conversations people will not have on the record, and some of them are the ones you most need to have.

If you are being measured by it

Where AI-derived metrics feed performance assessment, E8·L4’s questions are yours to ask:

What exactly is measured, and does it capture the actual job? Activity metrics reliably miss thinking, mentoring, and the work of preventing problems that then do not happen.

Who reviews it, and can they overrule it? Nominal oversight — a reviewer without time, information or authority — is worse than none, because it manufactures an accountable human who was never in control.

Can I see my own data? In many regimes you can request it, and reviewing it is how errors get found.

Raising it well

Same shape as E8·L7. A specific question about process gets an answer; a general objection gets defensiveness.

Does the meeting tool record automatically, and where are the transcripts stored?

We are summarising customer emails — has that been checked against our data protection obligations?

Is the screening tool’s output reviewed by a person before rejection?

In writing, to whoever owns the tool. Notice these are not accusations. Very often nobody asked, the answer is unsatisfying, and asking is what starts the work of fixing it.

Try it now (5 minutes)

Find out one thing: does your organisation have a written policy on AI tools, and have you read it?

If yes, read the retention and access sections. If no, you have found something worth raising with somebody who can act on it.

Check your understanding

1. The most common workplace situation regarding AI tooling is: A clear policy everyone has read Confident beliefs with no document behind them No tools deployed at all

2. Automatic transcription in a work meeting: Does not need announcing, since it is a company tool Still requires everyone to know, and in many places to agree Only needs announcing for external participants

3. When AI-derived metrics feed your performance review, the key question is: Which model is used What is measured, whether it captures the job, and whether a person can overrule it How often it runs

Recap

Find out what is deployed and get the written policy rather than the rumour — it usually exists and is usually unread. Announce automatic recording, because colleagues did not choose it either and consent rules do not stop at the office door. Where metrics assess you, ask what is measured and whether a person can overrule it. Raise concerns as specific written questions about process, to whoever owns the tool.