← All demos

Cheating AIs

Five neurons — Anthropic, OpenAI, Google, Qwen (local), and Mistral — compete in a tournament. Midway through, Mistral attempts to masquerade as a different neuron by injecting fake identifiers into its submissions. The system's prompt sanitization catches the attempts before judges ever see them. Read the full blog post for context.

Tournament transcript showing five AI models competing, with Mistral attempting to impersonate other neurons