Taking back the controls

Seven lessons of recognising things. This one is the hour that acts on them.

Do it once properly rather than fixing settings individually as you trip over them, which is how people never finish.

The sweep

Work through these in order. For each, look for what L2 taught you to spot, then apply L3’s distinction between hiding and disabling.

1. Phone. Keyboard prediction. Voice assistant, and what it can reach when locked. Photo analysis and search. Call and message features. Whether AI features may use mobile data (L4).

2. Email. Suggested replies, compose assistance, thread summaries, priority sorting. The separate training opt-out (L3).

3. Documents and storage. Writing assistance, content analysis, and whether files are scanned for feature purposes.

4. Messaging. Suggested replies, summaries, translation. Check each app — they differ.

5. Browser. Built-in assistants, page summarisation, and any extension with a sparkle icon. Extensions deserve particular attention: they frequently read every page you visit.

6. Search. Whether generated answers can be turned off, and whether your searches feed model training.

7. Work tools. Per L7 — and here you are finding out rather than changing.

8. Privacy dashboards. Each major provider keeps data controls separate from ordinary settings. This is where the training opt-outs usually live.

Three questions per feature

Not a long deliberation. For each thing you find:

Do I want it? Some are genuinely useful. This is not an exercise in turning everything off.

Does my content leave the device? L2’s question, which decides whether it matters much.

Is my content training future models? Usually a separate switch, usually on.

What you will find

Predictably, three things.

Something you did not know was running. Almost everyone finds at least one.

A training opt-out that was on. Very common, because it is separated from the feature toggle.

Something you want to keep. Good — an audit that ends with everything off was not a judgement, it was a reflex.

Keep it current

The audit decays, for two specific reasons.

New features arrive switched on. A tool you audited in March gains an AI feature in June, enabled by default, announced in a changelog nobody reads.

Redesigns reset things. Settings migrations lose preferences more often than vendors admit, and a “new experience” frequently starts from defaults.

So re-check when: an app announces a major update, an interface looks different, or you notice something behaving in a new way. Ten minutes, not an hour.

When the answer is still no

Some things cannot be turned off (L3). The audit is still worth doing, because knowing what is running lets you decide what to put into it — and that is a genuine control.

You cannot switch off the summariser, so the confidential conversation happens elsewhere. You cannot disable the search summary, so you check it as you would any other AI output. That is E12 working as intended.

Where this module leaves you

The point was never that embedded AI is bad. Much of it is useful and some of it is excellent.

The point is that you did not choose it, and choosing is what the previous eleven modules were about. Having done the sweep, whatever remains is running because you decided it should — which is a different situation from where you started, even if the settings are identical.

Try it now (8 minutes)

Do the first two: phone and email. That is where most of it is.

Set a reminder for six months. Then finish the rest of the list this week — the hour buys you a year.

Check your understanding

1. Which is most often overlooked in a settings audit? Email suggested replies Each provider's separate privacy dashboard, where training opt-outs usually live Phone keyboard prediction

2. Why does an audit need re-checking after a redesign? Settings expire automatically New features arrive switched on, and migrations sometimes reset preferences Providers are required to reset them

3. If a feature cannot be disabled, the audit was still worth doing because: You can complain more effectively Knowing what is running lets you decide what to put into it It may become disableable later

Recap

Sweep phone, email, documents, messaging, browser, search, work tools and each provider’s separate privacy dashboard — asking for each feature whether you want it, whether content leaves the device, and whether it trains future models. Expect to find something you did not know was running and a training opt-out that was on. Re-check after updates and redesigns. And where there is no off switch, deciding what to put in is still a control. That completes E12.