Google launches a pilot of double-blind AI evaluations, keeping external evaluations in a cryptographic "box" to stop benchmark contamination and protect IP (Google DeepMind)

Google’s DeepMind team has begun a pilot that conducts double-blind AI model evaluations, placing external assessors and their results inside a cryptographically sealed “box.” This approach prevents benchmark contamination and safeguards proprietary intellectual property, aiming to create trustworthy, tamper-proof performance measurements for closed-source models.