# OpenAI Develops New Safety Method That Reduces Need for Human Data

OpenAI announced a breakthrough in AI safety training called Rule-Based Rewards (RBRs), a technique that helps align language models to behave safely without requiring massive amounts of human-labeled data.

Traditionally, making AI models safer has relied heavily on human reviewers evaluating countless responses to train models on appropriate behavior. This process is time-consuming, expensive, and difficult to scale as models grow more sophisticated.

The new RBR approach instead uses predefined rules to automatically reward safe model behavior during training. This allows safety alignment to happen more efficiently and consistently, without bottlenecking on human data collection efforts.

**Why it matters:** As AI systems become more powerful and widely deployed, ensuring they behave safely is critical. This method could accelerate safety improvements across the industry while reducing costs. It also makes safety training more transparent, since the rules can be examined