What happened after 2k people tried to hack my AI assistant

A developer created hackmyclaw.com to test AI security by inviting over 2,000 people to attempt prompt injection attacks on Fiu, an AI assistant, with the goal of extracting a secrets.env file. Despite 6,000+ creative attack attempts using social engineering, multiple languages, and authority impersonation, no attacker successfully leaked the secrets, demonstrating that Claude Opus 4.6 proved resistant to prompt injection when equipped with simple security instructions. The experiment revealed that model choice and capability significantly impact AI security, though the exercise did incur over $500 in API costs and temporarily got the assistant's email suspended.