The most recent generative AI fashions usually are not simply stand-alone text-generating chatbots—as a substitute, they will simply be hooked as much as your information to offer personalised solutions to your questions. OpenAI’s ChatGPT can be linked to your Gmail inbox, allowed to examine your GitHub code, or discover appointments in your Microsoft calendar. However these connections have the potential to be abused—and researchers have proven it could possibly take only a single “poisoned” doc to take action.
New findings from safety researchers Michael Bargury and Tamir Ishay Sharbat, revealed on the Black Hat hacker convention in Las Vegas at this time, present how a weak point in OpenAI’s Connectors allowed delicate data to be extracted from a Google Drive account utilizing an indirect prompt injection attack. In an illustration of the assault, dubbed AgentFlayer, Bargury reveals the way it was doable to extract developer secrets and techniques, within the type of API keys, that had been saved in an illustration Drive account.
The vulnerability highlights how connecting AI fashions to exterior programs and sharing extra information throughout them will increase the potential assault floor for malicious hackers and doubtlessly multiplies the methods the place vulnerabilities could also be launched.
“There’s nothing the person must do to be compromised, and there’s nothing the person must do for the info to exit,” Bargury, the CTO at safety agency Zenity, tells WIRED. “We’ve proven that is utterly zero-click; we simply want your e-mail, we share the doc with you, and that’s it. So sure, that is very, very unhealthy,” Bargury says.
OpenAI didn’t instantly reply to WIRED’s request for remark in regards to the vulnerability in Connectors. The corporate launched Connectors for ChatGPT as a beta characteristic earlier this yr, and its website lists at the least 17 completely different companies that may be linked up with its accounts. It says the system lets you “carry your instruments and information into ChatGPT” and “search recordsdata, pull reside information, and reference content material proper within the chat.”
Bargury says he reported the findings to OpenAI earlier this yr and that the corporate shortly launched mitigations to stop the approach he used to extract information through Connectors. The way in which the assault works means solely a restricted quantity of information could possibly be extracted directly—full paperwork couldn’t be eliminated as a part of the assault.
“Whereas this subject isn’t particular to Google, it illustrates why growing sturdy protections in opposition to immediate injection assaults is vital,” says Andy Wen, senior director of safety product administration at Google Workspace, pointing to the corporate’s recently enhanced AI security measures.
