A sharing feature in Meta’s new artificial intelligence (AI) platform, Muse, left thousands of users’ private conversations accessible to the public, according to an Oct 18 report from an Internet security firm.
Researchers with Claroc discovered that a feature that generates a link for users to send to others was built in a way that allowed that link to be searchable on the Internet. The link opens a new webpage that doesn’t require a login to access.
Though Meta took steps to make sure web crawlers and search engines couldn’t index those links, like in similar leaks at OpenAI, Anthropic and xAI, a conversation naming feature on the link and a common domain for the pages allowed the links to be searchable.
The firm wrote in the report that it found more than 6,800 links online. Though it said most conversations were “harmless,” some included personal records, internal business documents, and even a user’s “confidential prep for an executive-search call with a frontier AI lab.”
“To anyone looking to cause harm, a list like that is a shortcut, ready-made for fraud, impersonation or social engineering,” Claroc researchers wrote.
The firm said it reported the findings to Meta on Oct 8 through its “Bug Bounty” program, which incentivises people to report potential security breaches to the company with cash rewards up to US$300,000 (RM1.22mil) – the larger the security issue found, the larger the payout. Claroc’s chief technology officer, Asaf Aprozper, told Ars Technica that Meta did not respond to its report through the program, and that Claroc even tried reaching out directly to Meta employees through LinkedIn without success.
Meta suspended the feature after Ars Technica reached out for comment when first reporting on the issue. Claroc researchers wrote that previously accessible links no longer opened as of Oct 20, according to an update to the report.
Meta spokesperson Tracy Clayton confirmed to SFGate that the sharing feature would be paused while it reviews Claroc’s report. The security firm did not respond to SFGate’s request for comment on whether it received or expected to receive any payout from Meta.
Meta launched Muse on Sept 8, and it’s become widely popular. It’s one of only a handful of apps to reach more than 10 million downloads within 40 days of being on the App Store, joining the ranks of Fortnite and Pokemon Go. Meta markets the tool as a personal AI assistant, capable of handling its users’ mundane tasks like booking flights, making restaurant reservations and facilitating purchases.
But the AI tool requires deep personal access to run these features, worrying privacy experts given Meta’s lacklustre record when it comes to securing personal data. Recently, a New Mexico judge ordered Meta to pay US$25bil (RM102.14bil) for misleading users about its data collection practices. Claroc’s researchers wrote in the report that although these AI tools are useful, users should be more cautious of their privacy.
“AI assistants are only going to get more capable and more woven into how we work, so the practical course is to use them with your eyes open: assume everything you share is public until you have checked, and make sure someone can see what leaves,” the report said. – SFGate, San Francisco/Tribune News Service
