OpenAI has pulled the plug on a short-lived change to the configuration of the ChatGPT app that allowed users to make their conversations accessible to search engines. The decision was made after it became apparent that some private or commercially sensitive material was inadvertently being made accessible on the internet.
Barry Scannell, an AI law and policy partner at William Fry and a member of the Government-appointed AI Advisory Council, said his “jaw hit the floor” when he saw some of the material made accessible to routine Google searches on Thursday.
Dane Stuckey, Open AI’s chief information security officer, later said the feature that allowed users to make their conversations accessible for indexing by search engines would be disabled by Friday. Mr Dane Stuckey described the original move as “a short-lived experiment”.
He said the company was working to ensure all information that had been indexed was entirely removed.
READ MORE
Mr Scannell said there had been widespread confusion initially as to how the information was becoming publicly accessible and whether all prompts to ChatGPT were impacted. It appears users were clicking a check-box that had the effect of making shared chats discoverable by search engines without them realising the consequences.
He said it was clear from much of the information that became accessible from user prompts that this was being done unintentionally.
“Based on what I’ve seen, some of the stuff was so personally sensitive and commercially sensitive that people clearly didn’t realise a random person could come along and do a simple search on Google and be able to find the chats.”
He said the issue did not appear to be a technical issue but rather highlighted the need for greater AI literacy on the part of users to better understand the tech they are using.
“People seem to have clicked a box to make their chats discoverable on a search engine, or make them indexable, apparently without understanding what that meant. It’s just people doing this without realising it.
[ US AI policy risks disaster because of TrumpOpens in new window ]
“What this shows, I think, is just how important it is to have critical thinking and AI literacy as a really key component of any national strategy dealing with it.”
Mr Scannell said the incident should serve as a warning to businesses as commercially sensitive material could have been exposed.
An extension of legal confidentiality protections might also be required, he suggested.
It is also a concern that deeply personal information could have been made available, including the contents of chats in which individuals were using ChatGPT for the purposes of therapy.
