Intrusive AI monitoring of workers

What it is

Employers have always monitored work to some degree. AI changes the scale and depth: it can read every message, transcribe every call, score tone and sentiment, track keystrokes and screen activity, and combine these into a judgement about each person. Tools bought for security or quality can do all of this by default. The monitoring becomes a risk when it is more than the purpose requires, when staff are not told, when it reaches into private matters, or when its outputs are used to manage people. Inferring emotion at work is prohibited outright in the EU, as covered in workplace and education emotion recognition.

How it happens

• A security or productivity tool is deployed with all its analytics switched on.
• Monitoring introduced for one purpose is reused for performance management.
• Staff are told in general terms, or not at all.
• No assessment was made of necessity and proportionality.
• Meeting and call transcription captures private conversation.

Why it matters

Monitoring of workers is processing of their personal data and must be lawful, fair, transparent and proportionate; the ICO has published specific guidance. Excessive monitoring harms wellbeing and may breach rights to privacy. Staff who feel watched disclose less, including the problems an organisation most needs to hear about.

Warning signs

• No data protection impact assessment covers the monitoring.
• Staff do not know what is collected or how it is used.
• Monitoring data is used in disciplinary or performance decisions it was not introduced for.
• Sentiment or behavioural scores are produced for individuals.

Who owns it

The executive responsible for people is accountable for how staff are monitored. The owner of each tool is accountable for its configuration. The data protection officer advises. Staff and their representatives are informed and heard.

In practice

A contact centre introduces a quality tool that scores every call for empathy and flags staff whose scores fall. The scores turn out to be lower for staff with certain accents, and they are being used in pay reviews. The same pattern generalises to any monitoring tool adopted for what it can measure before anyone asked what should be measured.