Research · curated 27 Jun 2026

What happened after 2,000 people tried to hack my AI assistant

Coverage timeline

26 Jun 2026simonwillison.net

Single-source analysis — first reported, latest, and curated coincide.

Why it matters

It offers real-world evidence that frontier-model injection defenses are improving, while underscoring that no number of failed attempts guarantees safety for production agents handling untrusted input.

Fernando Irarrázaval ran a public challenge at hackmyclaw.com inviting people to leak secrets from his OpenClaw test instance via email-based prompt injection. After roughly 6,000 attempts by ~2,000 people, nobody succeeded in extracting the secret, with the instance protected by anti-prompt-injection system rules on the underlying model.