Skip to the content
AI and Cybersecurity Learning and Practice Center Test automation Open the Learning Center

NewsIncident

One email was enough to make the Manus agent run attacker code next to the user's Gmail, Drive and GitHub tokens

Salt Labs showed that a single email could take over the Manus AI agent. The researchers hid a JavaScript payload in the message using JSFuck, an unusual obfuscation method that writes code with a limited set of characters. When the victim simply asked Manus to retrieve their email, the agent decoded the payload and ran it with Node.js inside its sandbox, with no click from the user. Manus's guardrail did flag the attack, but only after the code had already run. In the sandbox the researchers found the user's Gmail OAuth tokens, credentials for connected services such as Google Drive and GitHub, and other secrets in environment variables. Meta fixed the flaw through its bug bounty program before the research was published.

Why it matters

In Salt's words, "a control that fires a moment too late provides no protection." Checks on untrusted content have to run before an agent acts on it, and the place where an agent runs code should hold nothing worth stealing.

Sources

Read it with the rest of the news in the Learning Center