Casey Newton

And I thought, well, let me see if I have any tools at my disposal that I can use.

And so I know that a lot of the AI image generators embed these digital watermarks in the image that are designed to be useful in exactly cases like this.

So I took the photo that he had sent me of his badge and I put it into ChatGPT and I said, does this raise any red flags for you?

The answer I got was no.

And then put it into Google Gemini, which came back and said,

this image was generated in whole or part with Gemini.

Gemini has what they call a synth ID, which is a watermark that's embedded in the image.

And so all of a sudden, I had something that looked pretty much like a smoking gun that I was being hoaxed.

It did.

And I do think it is just something that good actors in this space should be doing because we want to have all the tools we can at our disposal to help us sort through things like this.

Well, one reason why I was initially inclined to believe this source was that in the past, it would have taken a monumental effort to generate 18 pages and a fake employee badge, right?

In fact, after I published my story, I heard from a reporter at NBC News who had also been messaging with the same source.

And she shared her employee badge with me from NBC.

And it was almost identical to the one that he had shared with me.

So we now believe that he had just taken her employee badge, which she had sort of shared with him as a kind of good faith effort to build trust, fed that into Gemini and said, turn this into an Uber Eats badge.

And within seconds, he had it.

So it is just that easy.

It did not used to be.

And I think we as reporters and people who are consuming the news need to kind of catch up.

Yeah, completely.