Hugging Face exec warns cyberattacks threaten to make open source AI a little less open | TechCrunch

Recent attacks raise concerns

Fireworks disclosed on Monday that attackers had accessed some customers’ model weights, datasets and billing information. The company temporarily disabled some self-service dataset functions while investigating; the attack was not necessarily caused by the company’s openness.

Open platforms face competing pressures

Solaiman said the same permissive approach that makes open communities valuable can attract attackers. She described a desire to let hobbyists without institutional affiliations contribute, while noting that some people can undermine that openness.

Platforms also hold information that can draw attackers. Solaiman said evaluation datasets made Hugging Face tempting to the agent swarm, while OpenCode’s third-party API keys were an attractive target.

Possible limits could exclude contributors

A security-driven pullback could involve restricting who may upload models or datasets, requiring stronger identity checks, or vetting developers before granting access to some features. Such steps might protect platforms but make participation harder for independent researchers and hobbyists.

Support for open AI models has come from industry figures including Nvidia’s Jensen Huang, while governments, enterprises and startups increasingly rely on them. Solaiman warned that open source could become less accessible despite growing public support.

Hugging Face tightens security

Hugging Face is increasing monitoring and taking steps to prevent attacks, Solaiman said, while noting that monitoring can be costly. She said the company now rotates administrative access tokens more often and has stricter network controls.

GitHub is also tightening security after attackers inserted malicious code into projects to spread malware. The company has strengthened requirements for publishing npm packages, revoked older authentication tokens and limited the validity of new ones.