Anthropic and OpenAI need independent safety evaluators, experts say in public letter (www.cnbc.com)

Over 100 AI experts have issued a public letter urging that independent safety evaluators be integrated into the assessment of frontier AI technologies. The signatories, which include notable figures such as Geoffrey Hinton and members from prestigious institutions, emphasize that current resources and protections are inadequate for effectively testing AI safety. They advocate for independent evaluators to have scientific objectivity, transparency, and robust protections from company influence to assess the rapid advancements and associated risks of AI systems. This initiative follows recent announcements from major players in the AI sphere, like Anthropic, highlighting the critical need for accountability in the burgeoning AI landscape. This call for independent oversight is significant as it seeks to establish standardized practices within the AI industry, ensuring that evaluations are credible and free from conflicts of interest. The experts highlight that access to internal systems and communications will greatly enhance evaluators' ability to identify potential risks. As the capabilities of AI models expand, concerns around security, economic stability, and public safety grow. The letter underscores the necessity of independent assessments to safeguard against the unchecked development of powerful AI technologies, reflecting a collective effort to foster a more responsible and transparent AI ecosystem.