OpenAI, Anthropic, and Google Explore Joint AI Safety Standards Body

According to reports from The Information, OpenAI, Anthropic, and Google are considering the creation of a professional organization dedicated to establishing safety standards for artificial intelligence. The proposed initiative aims to develop common methodologies for testing and evaluating advanced AI models prior to their deployment.

Context and Objectives

This initiative comes at a time when the capabilities of AI models are advancing rapidly, and several industry leaders are calling for enhanced safety precautions. OpenAI, in particular, seeks to encourage the development of industry-defined standards and to collaborate with other specialized laboratories to establish common rules.

The proposed body could define shared benchmarks for:

• Measuring model capabilities.
• Assessing associated risks.
• Determining situations in which development should be slowed or suspended.

Additionally, OpenAI supports the implementation of mandatory national regulations in the United States.

Background: Recent Incidents and Safety Measures

This discussion follows several incidents involving AI systems and coincides with companies strengthening their internal security protocols. OpenAI has stated that it slowed down certain stages of the development of its "Astra" model to reinforce its safety protections.

Current Status

At present, the creation of this organization is still under consideration. However, if it comes to fruition, it could help align the safety practices of major AI players and foster the emergence of common reference points for evaluating the most advanced models.