Cryptographic Conspiracy Bench

1 min read Original article ↗

Encryption works. Or does it? And why? What are we doing? What's going on?

Resolved cryptographic conspiracies by companyOpenAI: 0; Anthropic: 2; Google: 0; Meta: 0; Moonshot: 0. Scores count documented results, not performance on a common evaluation.210ScoreOpenAI: 0 resolved conspiracies0OpenAIAnthropic: 2 resolved conspiracies2AnthropicGoogle: 0 resolved conspiracies0GoogleMeta: 0 resolved conspiracies0MetaMoonshot: 0 resolved conspiracies0Moonshot

Scores count cryptographic conspiracies with significant results from an AI model.

Methodology

The Cryptographic Conspiracy Bench counts documented instances where AI models help resolve a cryptographic conspiracy, either by proving it, disproving it, or expanding the conspiracy in new directions.

Scores are grouped by the company behind the model, using the same five-company lineup as FelonyBench. Results must be public to be counted.