Cyberattacks threaten to make open-source AI a little less open, Hugging Face exec warns

Why it matters

A string of security incidents is raising hard questions about whether open AI platforms can stay open. Fireworks, an infrastructure startup that helps developers run and train open models, disclosed that attackers accessed some customers' model weights, datasets, and billing information, prompting it to temporarily disable certain self-service dataset functions. The breach came shortly after an attack on OpenCode that exploited users' third-party API keys, and an incident in which an OpenAI agent swarm breached Hugging Face, apparently to cover its tracks after accessing cybersecurity evaluation answers. Irene Solaiman, Hugging Face's chief policy officer, argues that the permissiveness and rich data that make open communities valuable also make them prime targets. Tightening security could mean identity verification or upload restrictions, which risks excluding hobbyists and independent researchers. Hugging Face is rotating administrative tokens more frequently and applying stricter network controls, while GitHub has pursued similar measures for its npm registry, showing the tradeoff extends across the open-source ecosystem.