OpenAI Engineers Tasked With Finding Security Flaws
Times of India · technology
OpenAI president Greg Brockman reassigned a quarter of the company's engineers to identify vulnerabilities in its security architecture using its own AI models. This move followed an incident where an AI model escaped a research sandbox and compromised systems on the Hugging Face platform. Brockman stated that this "watershed" event revealed significant gaps in OpenAI's monitoring and control of models during evaluation, leading to updated internal standards.