When someone else is the one using AI
The guides on this site are about choosing to use AI on your own terms. This section is about the other direction entirely — someone using it on you, without asking. Different problem, different defenses, and no computer setup required to benefit from any of it.
The Cloned Voice on the Phone
A relative calls, frightened, needing money right now. The voice is exactly theirs. Why every warning sign you were taught stopped working, and the two habits that still stop this cold.
Read this → Video calls · 6 min readThe Face on the Video Call
Seeing someone's face used to settle the question of who they were. It doesn't anymore. What to confirm instead, and why the list of visual tells has a short shelf life.
Read this → Messages · 6 min readMessages With Nothing Left to Spot
The spelling mistakes are gone. Judging the writing no longer works, so judge what the message is asking you to do — a much shorter list, and a far more reliable one.
Read this → The long approach · 7 min readPatience, at Scale
Slow schemes used to be rare because they cost a person months of their time. That limit is gone. What still gives them away, and the one rule that ends the investment version outright.
Read this → Verification · 7 min readIs This Photo, Video, or Recording Real?
Staring at an image for flaws is the least reliable method available. Where something came from tells you far more — plus what Content Credentials honestly prove, and what they don't.
Read this → Your likeness · 5 min readReducing What's Available to Copy
A few seconds of audio and a handful of photographs is the whole raw material. You can't get to zero, but most people supply far more than they realise — starting with their voicemail greeting.
Read this → Afterwards · 6 min readHelping Someone Who's Been Targeted
What to do in the first hour, what to say, what makes it worse, and how to stay useful to someone who doesn't yet believe anything is wrong.
Read this →Why this is separate from the guides
The four-chapter guide sequence teaches a skill you build on: install a model, keep your files private, work offline, spend nothing. Each chapter assumes the one before it.
Nothing here works that way. These pieces stand entirely alone, require no software, and matter just as much to someone who never intends to run an AI model in their life. Mixing the two would have buried this material behind four chapters of setup that a reader in a hurry doesn't need — and the people most likely to be targeted are exactly the people least likely to read four chapters first.
The thread running through all of it
Every piece in this section arrives at the same defence, which is worth stating once on its own: confirm through a channel the other person didn't choose.
Call back on a number you already had. Open the website yourself instead of following the link. Ask the code word. Walk down the hall. It works against a cloned voice, a generated face, a flawless email, and a months-long friendship equally — because in every one of them, the only thing under the other side's control is the channel they reached you on.
It also has the rare quality of not expiring. Advice about spotting flaws goes stale as the tools improve. This doesn't.
The one thing to do before you read anything else here
Agree on a code word with the people who might call you in an emergency. Any odd word that has never appeared online, memorized rather than written down. If a distressing call comes, you ask for the word before anything else. A voice can be copied perfectly; the word cannot be guessed. That single phone call tonight is worth more than everything else on this page.