Independent safety evaluations are an effective way to prevent AI control failures

Artificial intelligence research, products, and impact across society.

The field of preventing harm from advanced AI systems, spanning alignment, evaluation, and decisions to slow or withhold models whose capabilities outpace their safeguards.

The practice of measuring and comparing the capabilities of AI models through standardized benchmarks and evaluation protocols.