Developer Fernando I. built hackmyclaw.com: anyone could email "Fiu," his OpenClaw-based AI assistant, and try to trick it into revealing the contents of a secrets.env file. Then it hit the Hacker News front page.
More than 6,000 emails from over 2,000 people poured in, every one a jailbreak attempt. The result: the secrets never leaked.
The real point
His motivation wasn't bragging rights. Modern AI assistants get access to your email, calendar, files, and the open web — so a successful prompt-injection attack isn't a party trick, it's a breach. The experiment is a useful public stress test of whether instruction-following guardrails actually hold under thousands of adversarial tries. This time they did.
Source: fernandoi.cl