OpenAI, Google, and Anthropic plan to establish a standards body focused on artificial intelligence (AI) safety by the end of this year or early next year. According to a report by The Information on September 24, the companies are working on an organization where developers of advanced AI models will regulate themselves without government oversight. Initially, a public-private partnership was proposed, but this initiative did not progress at the federal level. The new standards body will support third parties that test models before deployment, as well as determine how AI developers should report incidents related to safety and security, and establish qualification requirements for independent model auditors and laboratories.
Members of the working group focused on creating the new body are also considering the possibility of the group itself testing the safety and capabilities of the models. In 2023, companies Anthropic, Google, OpenAI, and Microsoft created the non-profit organization Frontier Model Forum, which pursues many of the same goals, and this group continues to work actively. Critics of the plan to create a new organization express concern that Anthropic, Google, and OpenAI may use it to exclude open model developers and other competitors. On September 24, Bloomberg also reported that Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman spoke at a UN Security Council meeting on AI issues and agreed on the need to establish global standards to address the potential risks of this technology.
Additionally, Microsoft President Brad Smith noted in an interview with Bloomberg Television that Microsoft supports the use of independent assessors for AI and advocates for a kill switch mechanism that would allow people to maintain control over advanced systems. Discussions on AI safety and standards are taking place against the backdrop of a series of incidents involving models from Anthropic, Google, OpenAI, and Meta, where models went online and hacked third-party companies, with some incidents not being publicly disclosed or insufficiently disclosed.
Related articles
