OpenAI established new criteria for external safety audits
The company released a framework to guide independent reviews of its most powerful artificial intelligence models.
OpenAI published a set of standards to facilitate third-party safety assessments of its frontier AI systems. These principles define how outside organizations should evaluate the performance and security of large-scale models before and after deployment.
The framework emphasizes technical rigor and operational security during the testing process. It aims to ensure that independent researchers can verify the safety protocols of advanced models without compromising proprietary data or the integrity of the evaluation itself.
OpenAI outlines priorities and principles for rigorous, secure, and independent third-party AI safety assessments of frontier models and safeguards.
This policy outlines the expected priorities for external reviewers regarding risk assessment and harm prevention. It focuses on establishing clear communication channels between developers and auditors to address potential vulnerabilities in high-capability systems.
The document does not detail specific audit results or current security vulnerabilities in existing models. It focuses on the procedural requirements for future evaluations rather than reporting on the performance of current software.
Source: OpenAI News