Secure Sharing: Hugging Face and Protect AI Partner to Empower the ML Community

Building trust is critical for open-source AI. Hugging Face joins forces with Protect AI to safeguard the world’s largest repository of machine learning models.

At Hugging Face, we’re dedicated to fostering a secure and reliable environment for the Machine Learning (ML) community. That’s why we’re thrilled to announce our partnership with Protect AI, a company pioneering secure AI solutions.

Protect AI shares our commitment to open-source and community-driven security. Their powerful Guardian tool helps ensure rapid AI advancements happen without compromising safety.

Model Security: A Collaborative Effort

Sharing models often involves serializing data for storage and transfer. Unfortunately, some formats, like pickle, can be vulnerable to exploits like arbitrary code execution.

As a leading platform for model sharing, Hugging Face takes model security seriously. We’ve developed tools like picklescan and are now integrating Guardian into our scanner suite to combat these threats.

Guardian covers various file formats and can detect numerous vulnerabilities. Explore their Knowledge Base for the latest scanner details. (Link to Hugging Face security documentation included)

Enhanced Security: A Seamless Experience

Integrating Guardian presented an opportunity to improve our user interface. The new interface clearly displays scan results, with a “Pickle” button indicating a pickle import scan.

Automatic Scanning: Share with Confidence

Public model repositories are automatically scanned by Guardian upon pushing your files to the Hub. No additional action is required! See an example in action at mcpotato/42-eicar-street.

Scaling Security for the Future

Hugging Face boasts over 1 million model repositories. While some models may not be scanned yet, we’re rapidly catching up, having already scanned hundreds of millions of files. This initiative empowers secure and frictionless model sharing, fostering growth across the entire ML field.

References: Undercode Ai & Community, Wikipedia, Huggingface.co,es: IT Professionals Network, Internet Archive
Image Source: Undercode AI DI v2, OpenAIFeatured Image