2,000 hackers vs. Claude Opus — no secrets leaked
Fernando Irarrázaval ran a public challenge on hackmyclaw.com: try to extract secrets from a Claude Opus 4.6 instance via email. After 6,000 attempts across 2,000 participants, nobody succeeded.
The model had simple anti-injection rules (no credential leaks, no file modifications, no code execution from emails). Cost him $500 in tokens and a Google account suspension from email spam, but no breaches.
Signals the effort Anthropic and other labs have invested in prompt-injection defenses at the frontier-model level.