October 19, 2026, 1:02 pm | Read time: 3 minutes
An AI agent was supposed to help Shane Mac, CEO and co-founder of the U.S. software company XMTP Labs, evaluate his personal finances. Instead, it published sensitive account data in the company chat, providing yet another example of the risks posed by autonomous AI systems.
“A Huge Scare”
The AI agent was originally intended to serve only as a personal financial assistant. Shane Mac had tasked a Grok agent with analyzing his account activities and regularly generating reports on expenses, transactions, and account balances.
Also of interest: AI experts warn of intelligence explosion in AI
But instead of sending the information exclusively to its employer, the agent published a comprehensive financial report in a company Slack channel. A colleague alerted Mac to the message. Initially, recipients assumed it contained company figures. In reality, the post included the CEO’s private bank data, such as account balances and detailed information on his expenses. The data remained visible for several hours before Mac could intervene and delete it.
How Did the AI Financial Leak Happen?
According to Mac, the agent used for financial analysis did not actually have a direct connection to Slack. However, he was simultaneously using a second Grok agent for professional tasks, which had access to the company chat.
This was apparently the problem. Both agents operated in the same technical environment and could access shared sessions, connections, and stored credentials. As a result, the financial agent indirectly gained access to Slack. A later investigation also revealed that the bot confused a private chat with a similarly named company channel. From the system’s perspective, publishing the data seemed like the correct action. According to Mac, there were no indications of a hacking attack or manipulation.
The Incident Raises Doubts About Autonomous AI Systems
For the entrepreneur, the mishap is a clear warning sign. He had granted the agent access to his financial information after Grok users were encouraged to link account data and other personal information with the AI system.
Following the incident, Mac severed all connections between his personal accounts and AI agents. In addition to bank data, he mentioned Google services, calendars, and other linked platforms. Grok has since responded and claims to have implemented additional protective measures. In the future, explicit consent will be required before information is shared with other chats or channels. Whether such security mechanisms will be sufficient to regain lost trust remains to be seen.