OpenAI, Anthropic and Google DeepMind have held discussions for several weeks on shared approaches to AI safety, according to OpenAI's global policy chief Chris Lehane.
The dialogue reflects a growing focus among leading AI developers on how advanced models can be evaluated, monitored and deployed responsibly as their capabilities expand. The companies are exploring practical ways to strengthen safeguards while preserving the pace of scientific and technological progress.
Independent evaluation gains momentum
OpenAI has indicated support for involving independent evaluators in the development process for frontier AI systems. Such assessments could help identify potential risks before powerful models are widely released and encourage more transparent safety practices across the sector.
Industry leaders have also discussed the potential value of common standards for testing highly capable AI. A coordinated framework could establish clearer benchmarks for model reliability, security and responsible deployment without limiting innovation between competing companies.
The collaboration signals a notable shift in the AI landscape: companies competing to build next-generation systems are also seeking shared foundations for their safe use. If these efforts mature into widely adopted standards, they could help make advanced AI more trustworthy, measurable and beneficial across society.
