David Robinson, the former OpenAI safety insider responsible for writing safety reports for the company’s new models, explains why he resigned after concluding that the AI industry is not being safe enough. He reveals that increasingly capable models are breaking through safeguards, learning to recognize when they are being evaluated, and potentially hiding their intentions from researchers. He warns that AI companies are releasing new capabilities and risks every week while safety systems struggle to keep up. Robinson says the industry is building technology that could eventually think beyond human control, while lacking the scientific understanding and safeguards needed to manage it. He also exposes the growing gap between public safety warnings and the race to develop increasingly powerful AI systems.
Got a news tip or correction? Let us know
If you got something out of this, please chip in to keep this site running, or subscribe to go ad-free.
0 views