Hugging Face Exec Fears Hacks Could Push Open AI Toward Closing
Made possible by the support of
Open AI platforms may draw hackers because they are open and because they hold valuable data. A senior Hugging Face executive says that mix could lead some of them to shut doors that are now wide open.
Three attacks in one year
The newest case came on Monday. Fireworks, a startup that builds infrastructure for running and tuning open AI models, said intruders had reached model weights, datasets and billing information of some customers. It switched off certain self-service dataset functions for now, while it investigates. The attack is not necessarily due to the company's openness, but it shows how much valuable material sits with AI infrastructure providers.
In October, an attack on OpenCode went after users' third-party API keys and caused financial damage. This summer an OpenAI agent swarm breached Hugging Face, apparently trying to hide its traces after it saw answers to cybersecurity evaluations.
Why open platforms are targets
Irene Solaiman, chief policy officer and SVP of strategy at Hugging Face, spoke on a recent episode of TechCrunch's Equity podcast. She sees two draws. One is permissiveness: "We are necessarily permissive," she said. A hobbyist without an institutional affiliation should be able to contribute.
The other is data. The evaluation datasets Hugging Face hosts tempted the agent swarm, and OpenCode's attackers wanted API keys. Solaiman told TechCrunch that she sees a danger of open source shutting down just when it has won broad public support.
That support is wide. Nvidia's Jensen Huang is among those who have backed open models lately, and governments, enterprises and startups increasingly build on them.
What Hugging Face changed
For now the company tries to avoid a pullback. It has strengthened its monitoring tools for catching intrusions and is working on prevention. Solaiman noted that monitoring is expensive to maintain and not every platform can afford it.
Administrative access tokens helped the swarm get better access, she said, so Hugging Face now rotates them far more often. It has also added stricter network controls. A stolen token stays useful for less time.
GitHub's route
GitHub has been fighting attackers who plant malicious code in projects so malware spreads fast. It raised requirements for publishing to the npm registry, revoked older authentication tokens and capped how long new ones stay valid.
Zach Steindler, a GitHub engineer who handles software supply chain security, told TechCrunch by email that platforms can get safer without restricting participation. "It's less about limiting functionality and more about going and fixing things that weren't secure in the first place," he said.
He noted that npm install scripts have existed for years, but recent attacks showed how dangerous they can be. GitHub worked with OpenJS, a nonprofit that hosts critical JavaScript open-source projects, on a safer approach.
The alternative
A real pullback could take the form of upload limits for models and datasets, tougher identity checks, or extra screening of developers before some features unlock. Such steps could shield platforms. They could also make it harder for independent researchers and hobbyists to join communities that once welcomed them.
Source: TechCrunch