Anthropic and Gray Swan host $300,000 prompt injection red-teaming challenge

Anthropic and Gray Swan are hosting a $300,000 indirect prompt injection red-teaming competition, sponsored by Anthropic, Meta, Google DeepMind, and Reflection AI. Anthropic engineer Boris Cherny says Anthropic has made significant progress on prompt injection for Claude models and frames the contest as a test of whether the problem is solved.

A new report describes the challenge as asking whether hidden text can trick a social media manager's AI into posting. Gray Swan says every participant receives a performance certificate, and top players may be invited to a network for paid red-team work. That report is marked as an advertisement.

The reports do not describe the competition's rules, entry requirements, or timeline.