> There's not much you can do about it, as sibling comment mentions it's a known gap. There is some work [0] in this space on the investigative side to trace the leak's source, but again the only way it would work is if you can obtain a leaked copy post hoc (leaked to press, discovered through some other means, etc.).

Those kinds of watermarks seem like they'd fail to a sophisticated actor. For instance, if that echomark-type of watermark becomes widespread. I supposed groups like the New York Times would update their procedures to not publish leaked documents verbatim or develop technology to scramble the watermark (e.g. reposition things subtly (again) and fix kerning issues).

With generative AI, the value of a photograph or document as proof is probably going to go down, so it probably won't be that big of an issue.

You could do really sneaky things like alter the space between words or other formatting tricks.