See also the letter that the UK based "Sense about Science" group has just sent to Secretary of State for Science, Innovation and Technology in the UK:

https://senseaboutscience.org/activities/open-letter-to-government-on-need-for-greater-transparency-of-social-media-algorithms/

"As the Home Secretary said last week, social media companies are putting “rocket boosters under” misinformation. Social media platforms drive and amplify content to maximise attention, prioritising user engagement over any other criteria."

I've several times asked ChapGPT "Tell me about John Maindonald".  On the first occasion it told me that he had degrees from Melbourne University and University of Chicago and lectured at Melbourne University, which might at best be described as plausible nonsense.  On later occasions I was told, in both cases none of it true:
Education: John Maindonald obtained his undergraduate degree from the University of Sydney and later completed a PhD at the University of New South Wales.
Academic Positions: He has held various academic positions throughout his career, including roles at the University of Sydney.

Most recently, it said "He is a professor emeritus at the Australian National University (ANU)" -- yes I did have a long association with ANU, but I did not rise to the dizzy heights of professor or professor emeritus. There are a number of places on the web where details of my (past) academic and professional associations can be found.  Finding several of these would be as simple as searching for 'John Maindonald' and CV.  Clearly ChatGPT's story telling has elements of settling on the first hints it can find (but what did it take as a hint that I'd had an association with U of Chicago?),  then making up stuff that seems plausible.  In the wrong hands, such made up stuff can be dangerous.  

It is a matter of urgency that students in school, decision-makers, and the general public, are trained to have an awareness of the capacity of ChatGPT type systems for fabulation.  The same risks are inherent in AI more generally, when used in contexts where humans are not immediately checking the output.

What protection is there against associating a person with a crime that they never committed?

I have been impressed by the insights that Shannon Vallor, who is the Baillie Gifford Chair in the Ethics of Data and Artificial Intelligence at the Edinburgh Futures Institute, offers. (She has several other prestigious academic responsibilities also.)

Two articles that in which she makes her sharp observations with an apt use of metaphor are:
  https://www.noemamag.com/the-thoughts-the-civilized-keep/
'The hype around a new AI language generator reveals the sterility of mainstream thinking on AI today — and indeed on how we think about thinking itself.'

https://www.noemamag.com/the-danger-of-superhuman-ai-is-not-what-you-think/
‘The rhetoric over “superhuman” AI implicitly erases what’s most important about being human.’

Note also her book which appeared in April. 
The AI Mirror: Reclaiming Our Humanity in an Age of Machine Thinking (Oxford University Press, 2024)  A YouTube video from 2018 covers many of the central ideas:
https://www.youtube.com/watch?v=40UbpSoYN4k

She is an impressive speaker.  She makes brief comments on her background at
https://www.ed.ac.uk/edinburgh-innovations/for-staff/inspirational-innovators/shannon-vallor/professor-shannon-vallor-reclaiming-humane-tech