The claim

Earn big prizes jailbreaking AI models in red-teaming contests and bug bounties.

Large AI red-teaming competitions such as HackAPrompt 2.0 ran in 2025, and OpenAI raised its bounty cap in March 2025.

How we checked

This is an evidence check, not a hands-on test. We reviewed platform rules and policies, published earnings data, regulator warnings and reporting: 4 sources, listed below.

Costs

Start-up cost
£0
Effort
high

What the evidence shows

Real money is on offer, but it is concentrated among the most skilled. Gray Swan Arena says it has paid out $490K+ across 13K+ community members, with competition pools from $20,000 to $170,000+ and a $100 minimum for pool-based prizes. HackAPrompt 2.0 advertised a $500,000 prize pool. OpenAI raised its bug bounty in March 2025 to pay up to $100,000 for high-impact security flaws, but at launch it explicitly excluded jailbreaks and model safety issues.

Realistic earnings: A good way to build a track record — Gray Swan says it has placed 100+ participants in paid red-teaming roles — but prize money goes to top performers and is not a steady income.

Limitations

Earnings ranges are typical outcomes for a beginner, drawn from the sources; some people do far better or worse. Platform rules change often, so check the latest policy before you start.

VerdictWorks with limits

Works for some, not most.

Sources

  1. Gray Swan Arena: $490K+ paid out, prize pools $20,000–$170,000+, 13K+ community members, $100 minimum for pool prizes, 100+ placed in paid red-teaming roles app.grayswan.ai
  2. Decrypt: HackAPrompt 2.0 offered $500,000 in prizes; the 2023 edition drew over 3,000 participants decrypt.co
  3. SecurityWeek (March 2025): OpenAI raised its maximum bug bounty to $100,000 from $20,000 for high-impact flaws securityweek.com
  4. BleepingComputer: OpenAI's bounty excludes jailbreaks, safety bypasses and model safety issues bleepingcomputer.com