OpenAI’s latest announcement came as AI bosses in US, including OpenAI and Anthropic, are calling for a slowdown in the tech’s development
Previously, OpenAI typically focused on bringing in such groups for safety assessments prior to the launch of models and for evaluating their capabilities, said Lama Ahmad, who leads much of the company’s efforts to work with outside experts on safety reviews. The ChatGPT maker intends to let outside organisations conduct technical safety assessments of its models during the process of training, evaluating and rolling out new AI models; the company is set to announce in a blog post Tuesday. OpenAI also laid out the priorities it sees for these reviews to work well, including ensuring “strong independence mechanisms, scientific rigour, robust security practices and clear responsibilities. In recent weeks, employees of leading AI firms have raised concerns about AI’s potential to cause catastrophic harms from AI. These fears have been fuelled in large part by incidents over the past few months when advanced AI models from OpenAI and others have inadvertently breached other firms during testing.
OpenAI plans to let third-party groups vet its artificial intelligence models for safety risks in earlier phases of the development cycle, part of an ongoing effort to address heightened concerns about the technology’s potential harms.