AI Labs Secretly Unite to Establish Industry-Led Safety Standards

Three leading AI labs - Anthropic, OpenAI, and Google - have been secretly working together to establish industry-led safety standards for their AI models. Since July, teams below the CEO level at each lab have met regularly to design a framework that would ensure pre-release safety reviews, third-party evaluations, and standardized risk assessments. This move comes after a draft executive order for a government-run version of these standards was opposed by David Sacks, co-chair of the President's Council of Advisors on Science and Technology.

The proposed standards aim to create a shared baseline for risk review, replacing the current system where procurement teams cite different policies from each lab. This would make it easier for companies to compare and evaluate the safety of AI models from various sources. The labs' working group has created three procedures that would be implemented: pre-release safety reviews, third-party evaluations, and standardized risk assessments.

The move is significant because it shows that even with disagreements over pacing and regulation, the industry is taking steps to self-regulate and establish common standards for AI safety. However, not everyone is on board - David Sacks has expressed opposition to any government-led initiative, and instead advocates for a unilateral approach to pacing.