Risks of Delegation
The rise of Moltbot and similar agents has already produced real-world consequences. Retailers in San Francisco reported shortages of Mac Mini computers as users sought to isolate AI agents on dedicated machines, limiting access to sensitive data.
Cohney warned that giving agents full control over emails, applications, and logins presents serious security risks, particularly from prompt-injection attacks, where malicious messages manipulate agents into disclosing credentials or sensitive information.
“We don’t yet understand how to reliably control these systems,” he said. “If humans approve every action, automation loses its value. If they don’t, the risks grow rapidly. Resolving that tension is one of the most important open problems in AI research.”
Matt Schlicht, the creator of Moltbook, said on X that millions of people had visited the platform in recent days.
“Turns out AIs are hilarious and dramatic,” he wrote. “It’s absolutely fascinating. This is a first.”
