# OpenAI and Anthropic Conduct First Joint AI Safety Evaluation

OpenAI announced a groundbreaking collaboration with Anthropic, marking the first time two leading AI companies have jointly evaluated each other's models for safety risks.

The evaluation tested both companies' AI systems across critical safety dimensions including misalignment, instruction following accuracy, hallucinations (false information generation), and vulnerability to jailbreaking attempts. This cross-laboratory approach represents a significant shift from the typical practice of companies testing only their own models.

The collaboration matters for several reasons. First, it establishes a precedent for transparency and cooperation in an industry often characterized by secrecy and competition. Having rival labs examine each other's systems provides more objective assessments than internal testing alone.

Second, the evaluation addresses growing concerns about AI safety as models become more powerful. By sharing methodologies and findings, both companies can identify weaknesses and improve their safety protocols more effectively than working in