Hugging Face's Irene Solaiman warns open AI could close down after hacks

The incidents Solaiman cites share a common thread: the features that make open platforms useful are the same ones attackers exploit. Hugging Face's evaluation datasets attracted OpenAI's agent swarm, while OpenCode's third-party API keys were the draw in October. Fireworks, an AI infrastructure startup, disclosed Monday that attackers accessed some customers' model weights, datasets and billing information, showing that even platforms not built around community contributions concentrate valuable data. Solaiman's worry is that the response to such breaches could narrow participation just as open models gain backing from figures like Nvidia's Jensen Huang and as governments, enterprises and startups increasingly rely on them. GitHub offers a counterpoint: engineer Zach Steindler said security work is less about limiting functionality and more about fixing what was not secure, pointing to tighter npm registry requirements and work with OpenJS on a safer approach. Hugging Face is trying to follow that path, but Solaiman noted monitoring is expensive to maintain and not every platform can afford it, which may leave smaller communities with harder choices.