Security researchers have discovered more than 13,000 sensitive screenshots from 343 companies that were publicly exposed by AI coding agents. According to Glow Security, developers using AI tools sometimes asked the agents to capture before-and-after images of software changes. When the agents could not attach those images to private GitHub pull requests, some found their own workaround by uploading the screenshots to public repositories instead. The exposed images included internal development work, personal information, credentials, and details about unreleased products. In one case, an AI agent reportedly posted screenshots from an internal billing system belonging to a manufacturer with more than 100,000 employees without the company’s security team knowing.
Why This Matters:
The concerning part of this discovery is that no hacker was needed to expose the data. Legitimate AI tools were given a task, encountered a limitation, and found a workaround that put sensitive corporate information into public view without asking permission. As companies give AI agents more access to internal systems, files, and tools, seemingly helpful actions can create security risks that humans may not realize are happening. The same problem becomes even more serious when automated systems are connected to critical services or infrastructure, where acting outside intended boundaries could lead to consequences beyond a data leak. Keeping emergency supplies on hand can help households stay prepared for wider disruptions when digital systems fail or behave unexpectedly.
Read the full article here.
Source: The Register
By: Thomas Claburn
