Endor Labs Launches Scoring System for AI Models: A Step Towards Secure AI Development
New Scoring System for Open-Source AI Models
Endor Labs has introduced a groundbreaking scoring system for artificial intelligence (AI) models, focusing on their security, popularity, quality, and activity. This initiative, known as ‘Endor Scores for AI Models,’ aims to simplify the identification of the most secure open-source AI models available on Hugging Face, a leading platform for sharing Large Language Models (LLMs) and machine learning tools.
Supporting Developers in AI Governance
The announcement comes at a time when developers are increasingly turning to platforms like Hugging Face for ready-made AI solutions, akin to the initial waves of open-source software (OSS). This new scoring system enhances AI governance by enabling developers to “start clean” with AI models, addressing a critical challenge in the current AI landscape.
A Commitment to Security in AI
Varun Badhwar, Co-Founder and CEO of Endor Labs, expressed, “It’s always been our mission to secure everything your code depends on, and AI models are the next great frontier in that critical task.” He highlights the growing experimentation with AI across various organizations, underscoring the importance of security in pace with this rapid evolution.
AI Models: A Double-Edged Sword
George Apostolopoulos, Founding Engineer at Endor Labs, noted that teams are exploring AI models for various purposes, from launching entirely new AI-driven businesses to enhancing existing products with AI capabilities. However, he warns that the current AI model landscape resembles “the wild west,” with developers often selecting models without assessing potential vulnerabilities.
A Comprehensive Approach to Risk Evaluation
Endor Labs approaches AI models as vital dependencies within the software supply chain. Apostolopoulos further elaborates, “Our mission at Endor Labs is to ‘secure everything your code depends on.’” This perspective enables organizations to apply established risk evaluation methodologies to AI models in the same way they would other open-source components.
Key Risk Areas Identified
Endor’s AI model evaluation tool targets several crucial risk areas:
- Security vulnerabilities: Pre-trained models may include harmful code or vulnerabilities that could lead to security breaches when integrated into an organization’s infrastructure.
- Legal and licensing issues: Compliance with licensing terms is essential, given the complex lineage of AI models and their training datasets.
- Operational risks: The reliance on pre-trained models results in a complex web of dependencies that can complicate management and security efforts.
Out-of-the-Box Checks for Enhanced Security
To address these concerns, Endor Labs’ evaluation tool applies 50 out-of-the-box checks to AI models on Hugging Face. The system generates an “Endor Score” influenced by various factors, including the number of maintainers, corporate sponsorship, release frequency, and any known vulnerabilities.
Positive and Negative Scoring Factors
The scoring system considers positive attributes, such as the use of safe weight formats, accessible licensing information, and robust download and engagement metrics. Conversely, factors like incomplete documentation and unsafe weight formats negatively impact a model’s score.
User-Friendly Interface for Developers
A standout feature of Endor Scores is its user-centric design. Developers are not required to know specific model names; they can initiate their search with broad questions like, “What models can I use to classify sentiments?” or “What are the most popular models from Meta?” This functionality streamlines the selection process, providing clear scores to guide developers in choosing the most suitable models.
Accelerating Innovation with Safe AI Models
“Your teams are being asked about AI every single day, and they’ll look for the models they can use to accelerate innovation,” Apostolopoulos remarked. Using Endor Labs to evaluate open-source AI models ensures that developers can trust the models they select to perform as expected and uphold security standards.
Conclusion: A New Era for AI Model Evaluation
As organizations venture into AI development, the tools and methodologies to ensure the security and compliance of AI models become increasingly crucial. Endor Labs’ scoring system represents a significant advancement in making AI model selection safer and more reliable, addressing both the immediate needs of developers and the broader implications for the software supply chain.
Frequently Asked Questions
1. What is the purpose of the Endor Scores for AI Models?
The Endor Scores aim to help developers identify secure, high-quality open-source AI models on Hugging Face by providing easy-to-understand scores based on security, popularity, quality, and activity.
2. How does the evaluation tool assess AI models?
Endor Labs’ tool applies 50 out-of-the-box checks, generating an “Endor Score” based on factors such as maintainers, corporate sponsorship, release frequency, and known vulnerabilities.
3. What are the main risk areas identified by Endor Labs?
The key risk areas include security vulnerabilities, legal and licensing issues, and operational risks associated with pre-trained models.
4. How can developers find suitable AI models using Endor Scores?
Developers can start by asking broad questions about model capabilities without needing to know specific names, allowing the tool to suggest appropriate models and their scores.
5. Why is AI model security important for organizations?
Ensuring the security of AI models helps organizations avoid potential breaches and legal issues, ultimately reducing maintenance costs and fostering trust in AI technologies.
