While AI agents promise massive productivity gains, cybersecurity experts warn that their biggest strength — broad system access — is also their greatest vulnerability. Researchers testing OpenClaw on Moltbook found agents highly susceptible to prompt injection attacks, where malicious instructions embedded in posts or emails trick agents into leaking credentials or transferring funds.

Security specialists from firms like Huntress and Permiso Security noted that agents connected to email, messaging apps, and financial tools could be manipulated with surprisingly little effort. Guardrails exist, but like humans falling for phishing scams, AI models can still be deceived.

For now, experts caution everyday users against deploying such systems widely. The productivity upside is enticing — even aligning with predictions from Sam Altman about solo founders building unicorns with AI — but without stronger security foundations, the agentic future may remain more hype than reality.