This popular AI agent could be hacked by a single email — with potentially disastrous consequences

Researchers found a way around Manus' guardrails and got it to execute a simple email prompt injection attack.