Microsoft patches AutoJack flaw in AutoGen Studio that let a malicious webpage run commands on the host
Developers running Microsoft's $MSFT AutoGen Studio should check their version. A flaw called AutoJack let a malicious webpage trick an AI agent into running commands on the host. The published release is safe, but two pre-release dev builds carried the bug and were not pulled. S
Developers experimenting with AI agents have a fresh reason to check what they are running locally. Microsoft $MSFT has disclosed and fixed a vulnerability called AutoJack in AutoGen Studio, its tool for building and testing AI agents. The flaw was a chain of three weaknesses. A local WebSocket interface trusted connections from localhost, the authentication layer left certain agent routes unchecked, and the interface accepted encoded parameters that could be used to run commands. Combined, they meant that simply visiting a malicious webpage could let an attacker push arbitrary PowerShell, Bash or executable commands onto the machine running the agent. There is important context on exposure. Microsoft remediated the issue before any general release, so the current published package, autogenstudio 0.4.2.2, is not affected. The catch is that the vulnerable handler did ship in two pre-release builds, 0.4.3.dev1 and 0.4.3.dev2, and those have not been pulled from the package index. Anyone who installed a development build should update. Microsoft's recommendation is to run AutoGen Studio under a low-privilege account inside a sandboxed profile or container, so that any future agent-driven command execution stays contained rather than reaching the wider system. Source: BleepingComputer, The Hacker News, TechRadar.