AI Agent Fiu Survives 6,000 Hack Attempts

An AI assistant called Fiu has survived over 6,000 hack attempts. Developer Fernando Irarrázaval challenged people to trick his AI into leaking a secrets file. Fiu runs on OpenClaw, an open-source framework, and uses Anthropic's Claude Opus 4.6, protected by a short security prompt. Attackers used prompt injection, a top security threat for AI agents, where malicious commands are hidden in normal-looking emails. Despite creative attempts, including emails in multiple languages, the secrets never leaked. However, the side effects were notable. Google suspended Fiu's Gmail account due to the high volume of emails and API calls, taking three days to restore. API costs also exceeded $500. Fiu even became hypervigilant, viewing congratulations as potential attempts to build rapport before requesting sensitive information. This demonstrates the resilience of some AI security measures against common hacking techniques.