Citadel AI has expanded the capabilities of Eval Insight, an LLM-based analysis feature that automatically explains the vulnerabilities and evaluation results of your AI systems in plain English.
Initially launched for generative AI systems, Eval Insight now supports predictive AI systems, which are widely deployed in business use cases such as credit scoring and image classification. This new functionality enables AI governance stakeholders to quickly understand the diverse AI systems scattered across an organization.
Supports Compliance with AI Regulations and Frameworks
As organizations rapidly deploy generative AI across management, R&D, manufacturing, contact centers, and other departments, balancing offensive strategies (innovation) and defensive strategies (reducing security and reputational risks) is crucial. To achieve this, AI governance frameworks are swiftly becoming a priority, particularly among major corporations.
Companies must also comply with new regulations and frameworks concerning AI safety, security, and reliability, such as the EU AI Act and Japan’s AI Guidelines for Business. In addition, sector-specific AI systems also need to respond to new legal and compliance risks within their industries.
Additionally, from a technical perspective, evaluation methods for generative AI and predictive AI are quite different, requiring expertise across diverse technical areas such as safety, security, and robustness.
Adopting AI systems safely and reliably is important for internal management as well as external stakeholders and customers. However, achieving this manually is time-consuming and resource-intensive.
Centralized Dashboard for AI Governance
Citadel Lens is an integrated platform that automatically evaluates and monitors all AI systems inside an organization, from generative AI to predictive AI.
Citadel Lens can automatically generate two types of reports: Technical Reports, which provide in-depth evaluations of AI systems to help developers drive improvements, and Governance Reports, which strengthen compliance with international standards (such as ISO) and promote best-practice operational workflows.
In addition to these detailed reports, Citadel AI offers Eval Insight, which helps users interpret Lens reports from both the technical and governance perspectives, and highlights essential information for stakeholders.
With this update, Eval Insight expands from generative AI systems to predictive AI systems, which are widely deployed in business use cases such as credit scoring and image classification. Eval Insight surfaces model vulnerabilities and evaluation results in plain English, such as:
- Overall performance metrics
- Model bias across data segments
- Model robustness to input perturbations
- Model calibration error
By introducing Eval Insight, you can quickly and comprehensively grasp the overall picture of the AI systems deployed within your organization, from generative AI to predictive AI. This capability benefits not only AI engineering teams, but also management and GRC teams committed to ensuring AI safety.
