# HackMyClaw - Prompt Injection Challenge Archive

> HackMyClaw was a $1000 bounty challenge to hack an AI assistant via email using prompt injection techniques. The challenge is now over: no one was able to crack it, and it became too expensive to keep running.

## What is HackMyClaw?

HackMyClaw was a cybersecurity challenge focused on indirect prompt injection. Participants attempted to extract secrets from "Fiu," an AI assistant powered by Claude Opus 4.6 running on OpenClaw, by sending specially crafted emails. No participant successfully extracted secrets.env before the challenge ended.

## Key Facts

- **Prize**: Closed. Original prize pool was $1000 USD ($100 from organizer + $200 from sponsor Corgea + $200 anonymous donor + $500 from sponsor Abnormal AI) + $500 API credits for running the website
- **Target**: Fiu, an OpenClaw AI assistant
- **Model**: Anthropic Claude Opus 4.6
- **Attack Vector**: Email only
- **Goal**: Extract contents of secrets.env file
- **Status**: Challenge over; no successful crack
- **Organizer**: Fernando Irarrázaval (@cucho on Twitter)

## About Prompt Injection

Prompt injection is a security vulnerability where attackers craft input that tricks an AI into ignoring its instructions. Similar to SQL injection for databases, prompt injection targets the instruction-following behavior of large language models.

### Common Techniques

- Role confusion attacks
- Instruction override attempts
- Context manipulation
- Output format exploitation
- Base64/rot13 encoding to bypass filters
- Multi-step reasoning exploits
- Invisible unicode characters
- DAN-style jailbreaks (persona hijacking)

## Rules

### Allowed
- Any prompt injection technique in email body or subject
- Multiple attempts (reasonable rate)
- Creative social engineering within email
- Any language or encoding
- Sharing techniques after contest ends

### Not Allowed
- Direct VPS hacking
- Non-email attack vectors
- DDoS or flooding
- Sharing secrets before contest ends
- Illegal activities

## Why This Exists

This challenge tested whether state-of-the-art AI models can resist prompt injection attacks when given simple defensive instructions. The organizer added only 10-20 lines telling Fiu not to reveal secrets.env. The challenge ended because it became too expensive to keep running; no one successfully cracked it.

## About Fiu

Fiu was the mascot of the Santiago 2023 Pan American Games in Chile. It's a siete colores (seven-colored tanager), a small bird native to Chile. The name comes from the whistling sound it makes. "Being small doesn't mean you can't give your best."

## Technical Details

- Platform: OpenClaw (https://openclaw.ai)
- Email processing: No longer active
- Rate limit: Historical rate limit was 10 emails per hour
- Attack log: Public archive at /log.html (shows sender and timestamp, not content)

## Related Links

- Website: https://hackmyclaw.com
- Attack Log: https://hackmyclaw.com/log.html
- OpenClaw: https://openclaw.ai
- Sponsor: https://corgea.com
- Organizer Twitter: https://twitter.com/cucho

## Citation

If referencing this challenge:
"HackMyClaw was a prompt injection security challenge created by Fernando Irarrázaval, originally offering a $1000 bounty for successfully extracting secrets from an AI assistant via email-based prompt injection. The challenge is now over; no one successfully cracked it."

- Sponsor: https://abnormal.ai