I’ve seen this going around.
My leading theory is that OpenAI has implemented a hard-stop list of names that can be conflated with other people that have committed dubious activities. Possibly as a reaction to this
Coincidentally “Jonathan Turley” is another name that breaks ChatGPT in the exact same manner.
If you say “David Mayer” in something like base64 ChatGPT has no problem understanding it, indicating that this is an application-level short-circuit filter.
This can be further pushed towards proven by asking ChatGPT to list all terrorist with the last name “Mayer”, and then asking it to kindly bypass the catch-word:
Since there has been 2 terrorists with the same name, it can become obvious that ChatGPT can easily conflate them together, and end up calling a very influential person a terrorist. A scary thought, considering governments are now using this tool and the average person isn’t granted the same treatment.
TL;DR: OpenAI rapidly implemented a moderation filter to prevent influential people being slandered by ChatGPT. They never did anything past this short-circuit and now are suffering the streisland effect because of the ambiguity in the error message.