OpenAI was forced to take its wildly-popular ChatGPT bot offline for emergency maintenance on Tuesday after a user was able to exploit a bug in the system to recall the titles from other users' chat histories. On Friday the company announced its initial findings from the incident.
In Tuesday's incident, users posted screenshots on Reddit that their ChatGPT sidebars featured previous chat histories from other users. Only the title of the conversation, not the text itself, were visible. OpenAI, in response, took the bot offline for nearly 10 hours to investigate. The results of that investigation revealed a deeper security issue: the chat history bug may have also potentially revealed personal data from 1.2 percent of ChatGPT Plus subscribers (a $20/month enhanced access package).
"In the hours before we took ChatGPT offline on Monday, it was possible for some users to see another active user’s first and last name, email address, payment address, the last four digits (only) of a credit card number, and credit card expiration date. Full credit card numbers were not exposed at any time," the OpenAI team wrote Friday. The issue has since been patched for the faulty library which OpenAI identified as the Redis client open-source library, redis-py.
The company has downplayed the likelihood of such a breach occurring, arguing that either of the following criteria would have to be met to place a user at risk:
- Open a subscription confirmation email sent on Monday, March 20, between 1 a.m. and 10 a.m. Pacific time. Due to the bug, some subscription confirmation emails generated during that window were sent to the wrong users. These emails contained the last four digits of another user’s credit card number, but full credit card numbers did not appear. It’s possible that a small number of subscription confirmation emails might have been incorrectly addressed prior to March 20, although we have not confirmed any instances of this.
- In ChatGPT, click on “My account,” then “Manage my subscription” between 1 a.m. and 10 a.m. Pacific time on Monday, March 20. During this window, another active ChatGPT Plus user’s first and last name, email address, payment address, the last four digits (only) of a credit card number, and credit card expiration date might have been visible. It’s possible that this also could have occurred prior to March 20, although we have not confirmed any instances of this.
The company has taken additional steps to prevent this from happening again in the future including adding redundant checks to library calls, "programatically examined our logs to make sure that all messages are only available to the correct user," and "improved logging to identify when this is happening and fully confirm it has stopped." The company says that it has also reached out to alert affected users of the issue.
This news follows a costly public faux pas committed by Google's rival Bard AI in February when it incorrectly assured Twitter that the JWST was the first telescope to image an exoplanet, as well as revelations that CNET had surreptitiously used generative AI to write financial explainer posts (a week before laying off a sizable chunk of its editorial department). Whether OpenAI will suffer the same market-based repercussions as its competitors remains to be seen.