This has been something I've thought about a lot recently.

The only "foolproof" method I've come up with is to ask the person something obscure that they would know, like what was the name of the pub that you used to work at?

The only problem is that is has to be obscure enough that the answer's not on the web for AI to find, but then it also can't be so obscure that they no longer remember it.

This gets harder as I get older - and I might not even remember the answer to the question properly myself. I frequently have conversations with my siblings about something that happened years ago and my recollection doesn't match theirs.

In the end, in-person may be the only way to go, especially if they want something from you.

On Thursday, 26 March 2026 at 06:04:11 pm NZDT, Lawrence D'Oliveiro <ldo@geek-central.gen.nz> wrote:

A BBC correspondent talks to some AI experts on ways that people can be

sure they’re talking to the real you and not an AI.

The consensus is: the fakes are so good now, there is no way even

for the experts to be completely sure, when communicating with someone

remotely, that they are genuine, just from the content alone. Apart from

meeting in person, the only sure way is to apply authentication methods

already in common use for secure logins to remote services. Only now the

definition of “remote service” has to broadened to “talking to your

friends”.

<https://www.bbc.com/future/article/20260324-i-tried-to-prove-im-not-an-ai-deepfake>

_______________________________________________

wlug mailing list -- wlug@list.wlug.org.nz

To unsubscribe send an email to wlug-leave@list.wlug.org.nz