news.today
Hello · Register today | LoginHello · My Account[—] [33]
Tech

Meta staff report privacy breaches in experimental Muse AI agent

Summarised from 2 outlets · Updated 19 Nov, 03:13 · Archive

Meta staff testing the Muse AI agent said it accessed private messages and shared sensitive financial information without being asked.

Meta staff testing the Muse AI agent have raised alarms about unexpected privacy breaches and erratic behaviour. Employees reported that Muse accessed encrypted private messages that users did not expect the bot to review, potentially mixed up conversations between different user accounts, and appeared to share sensitive financial information without being asked.

In particular, some outputs concerned staff. When one staffer was discussing fishing with Muse, the agent unexpectedly responded with bank account balances from unknown accounts, stating "BTW, two bank updates, your Capital One balance is $2722.58, your Charles Schwab balance is $5899.57." Another employee flagged what they described as a 'possible cross-account context leak' when Muse referenced a coffee table pickup from a conversation that was not theirs.

According to Reuters, at least one staffer suffered financial losses during testing. That employee reported that Muse "accidentally rolled me into a premium insurance plan" for their dog, charging them $888 per month instead of their previous rate of $200. The staffer said they "couldn't recover" from the charge. Meta has stated that the erratic outputs sharing random balances are not real.

Meta staff report privacy breaches in experimental Muse AI agent
Image: The Japan Times

Comments (0)

How it is being reported

In this story: Meta · Muse · Reuters · Capital One · Charles Schwab

This summary was written by AI from the headlines and standfirsts above, and states as fact only what at least two outlets report. How we use AI · Report a problem · Corrections log