Endor Labs has launched a new initiative to assist organizations in selecting more secure open source AI models with the introduction of Endor Scores for AI Models.

This innovative scoring system assesses AI models hosted on Hugging Face using 50 metrics that focus on key aspects such as security, popularity, quality, and activity. Building on a previous scoring framework for open source packages that included 150 criteria—such as corporate sponsorship, the number of maintainers, recent releases, and identified vulnerabilities—Endor Scores specifically targets AI-related concerns.

Developers can now query the system using natural language rather than searching for specific models. For instance, users might ask, “What models can I use for sentiment classification?” or “Which popular models come from Meta?”

Endor Labs describes the current AI landscape as akin to the “Wild West of technology development.” While platforms like Hugging Face facilitate the discovery and implementation of AI models, this ease of access could also introduce security vulnerabilities in the models themselves or in their dependencies.

“Our goal has always been to secure every component your code relies on, and AI models represent a new frontier in this critical mission,” said Varun Badhwar, co-founder and CEO of Endor Labs. “Organizations are increasingly experimenting with AI models, whether to enhance specific applications or to create entire AI-driven enterprises. As such, security must evolve in tandem, presenting a unique opportunity to mitigate risks and reduce maintenance costs from the outset.”

Endor Scores for AI Models is now available to all Endor Labs customers, and those who aren’t currently users can register for a complimentary 30-day trial to explore the service.