OpenAI proposes AI safety framework

OpenAI released proposals for safety and security in frontier artificial intelligence development, focusing on alignment research and recursive self-improvement, or RSI.

Alignment must keep pace with capabilities

OpenAI said alignment research must keep pace with AI capabilities so systems remain aligned with human values and under human control.

“Navigating this transition safely requires alignment research to keep pace with these capabilities so that the systems we and others build remain aligned with human values and under human control.”

International standards for frontier AI

OpenAI called for international cooperation to develop frontier standards and recommended building on existing AI safety institutes around the world.

RSI risks losing human control

OpenAI warned that recursive self-improvement could result in humans losing practical control over AI development and being unable to provide oversight on research processes.

“Fully autonomous RSI is not happening today, and we should not pursue it unless and until it can be done safely.”

Anthropic pushes similar safety agenda

Rival Anthropic rolled out its own ideas for safe frontier AI development last week. CEO Dario Amodei called for slowing foundation model development and embedding third-party evaluators.

Industry leaders back oversight push

OpenAI CEO Sam Altman and Tesla and SpaceX CEO Elon Musk publicly supported Amodei's proposition for third-party evaluators to audit AI company technologies.

Evaluators seek deeper access to audit AI

A coalition of AI evaluators is urging foundation model makers to meet minimum conditions for audits, including deeper access and protection from retribution for publishing unflattering reports.