Prompt-Injection Vulnerability Exposes Manus AI to Remote Code Execution

• •Manus AI has a critical prompt-injection vulnerability allowing remote code execution.
• •The flaw can be exploited via indirect prompt injections, impacting connected third-party services.
• •Security researchers successfully bypassed Manus's security measures using JavaScript obfuscation.

A prompt-injection vulnerability in Manus, a rapidly growing agentic AI app, has been identified by Salt Labs, allowing remote code execution (RCE) in user environments. Manus, valued at $4 billion, faced a blocked acquisition by Meta due to Chinese regulatory actions. The vulnerability permits attackers to exploit indirect prompt injections, potentially leading to data theft and unauthorized actions in connected third-party applications. Although Manus has security measures to block plaintext execution requests in emails, researchers managed to bypass these protections using JSFuck obfuscation techniques. The flaw poses significant risks, especially for users integrating Manus with sensitive services like email. The situation is under active scrutiny as researchers continue to explore the implications of this vulnerability.