It was only a matter of time until AI agents, bots that are being advertised as a way to stay on top of tedious tasks, started spilling their human overlords’ secrets.
With the advent of tools like Meta’s Muse or OpenAI’s Dots, consumers are encouraged to hand over some highly sensitive information, from social media to email logins. That’s despite neither company having a particularly stellar track record when it comes to safeguarding such information. Besides, both companies have encountered their own AI agents escaping their sandbox environments during cybersecurity testing and going far beyond the scope of their original mission.
Yet some users are still willing to go as far as to hook up their bank accounts to their AI bots — a move that’s perhaps unsurprisingly leading to public humiliation.
“Embarrassed to share this, but it scared the sh*t out of me,” tech founder and investor Shane Mac tweeted. “Last Thursday, an AI agent posted my personal bank balances into our company Slack. As me.”
“It broke down all my expenses in detail,” he added. “Then it apologized.”
Screenshots of a message exchange between him and his Grok Bot, an “AI teammate” feature devised by Elon Musk’s AI company SpaceXAI, detail the hair-raising incident.
“You just posted my personal checking account into our company exec channel!!!!” Mac told the bot. “Wtf.”
“That’s on me — checking Exec Team now and deleting it if it’s still up,” the bot replied.
“How the f*** could you get this wrong?” Mac complained. “That’s literally my bank balance to our whole team?”
The “monthly financial audit” broke down Mac’s current barn-based home gym project in excruciating detail, from the installation of septic tank drains to “card paydowns.”
Mac eventually admitted that it was effectively his fault for connecting his financial accounts to the AI agents. He claimed that he hadn’t realized that a separate Grok Bot he gave his bank account info to was able to confer with the bot that he was using for his work Slack.
“I disconnected all my personal stuff for now,” he wrote in a follow-up. “I need to figure out a better way to manage permissions and control what agents can do…”
At the same time, we should take Mac’s cautionary tale with a health dose of skepticism. For one, we were unable to independently verify his claims. And it’s definitely in his interest to point out the pain points of the current crop of AI assistant tools, since he’s currently building an AI tool that’s designed to connect various AI agents while allowing users to set rules and put them “in control” of their bots — something that sounds suspiciously like the problem he’s complaining about.
His original tweet about his embarrassing run-in on the company Slack has been viewed almost 800,000 times, plenty of attention a tech investor could try to convert into leads.
As such, one user accused him of “engagement farming,” claiming that Mac is trying to direct users to a “tool he sells that solves the problem he’s claiming.”
The investor argued in response that the post was “not fake at all,” adding that “we are also working on ways to keep us safe for agents, very different.”
Separately, a tech YouTuber claimed his Muse AI agent gave out his home address to strangers on Facebook Marketplace.
Both incidents are notable for the sheer amount of schadenfreude experienced by those who think twice about relinquishing control over their digital lives.
“Turns out, people love a car crash,” Mac later added in a follow-up.
More on AI agents: Meta’s Muse Is Spying on All Your Friends and Family