Citadel AI Launches a Next-Generation Firewall for Generative AI

Citadel AI announces the Lens Custom Firewall, an advanced firewall solution designed for generative AI applications. The Lens Custom Firewall precisely filters LLM input and output based on domain-specific rules, achieving safety comparable to continuous human supervision, even in real-time operations. The product aims to safeguard organizations from AI security and reputational risks, enhancing their AI risk management capabilities.

Challenges in Generative AI Security

As corporate adoption of AI accelerates, organizations face increasing pressure to balance the aggressive pace of innovation with comprehensive security measures. The use of LLMs in a corporate environment often requires guardrails to prevent harmful outputs, malicious attacks, and leaks of sensitive personal information. However, generic, off-the-shelf filters often fall short in real-world use cases, especially when applications require nuanced responses based on internal policies, customer databases, or industry-specific regulations.

Traditional content filtering often leads to overblocking (false positives) or underblocking (missed threats), compromising both security and usability. Moreover, generic tools struggle with context-sensitive domains like compliance or healthcare, especially in non-English languages, where precise decisions are crucial for maintaining quality, safety, and regulatory compliance.

Customizable, Next-Generation Firewall for AI

Citadel AI’s Lens platform enables businesses to easily create and fine-tune custom metrics tailored to their specific AI use cases. Rather than relying on generic, off-the-shelf metrics, Lens aligns closely with the nuanced insights of domain experts, significantly enhancing the precision of automated evaluation.

The newly launched Lens Custom Firewall further expands this capability into real-time operations. Leveraging Citadel AI’s proprietary “Metric Generator,” the solution automates the creation of custom metrics—drastically reducing manual workload during the evaluation phase. 

These custom metrics are seamlessly integrated into runtime firewall filters, continuously providing security comparable to expert oversight. This approach significantly reduces overblocking and underblocking issues, ensuring safe and trusted AI operations.

The Lens Custom Firewall precisely identifies and mitigates AI-related security and reputation risks according to each unique business context. This customized protection allows companies to manage risks optimally, balancing innovation and security to strengthen overall corporate resilience.

Key Features of Lens Custom Firewall

• Introduces Citadel AI’s proprietary Metric Generator to automate the creation of custom metrics
• Seamlessly integrates custom metrics and firewall filters, unifying evaluation and guardrail tooling
• Detailed customization of guardrail rules from the Lens UI
• Comprehensive logging and retrospective analysis of filtering decisions with clear, plain-English explanations
• Flexible control over latency requirements (asynchronous/synchronous filters, LLM-based/logic-based filters)

Related Information

• Citadel AI Advances AI Risk Management with Eval Insight

About Citadel AI, Inc.

Citadel AI provides software products that evaluate, monitor, and govern AI systems. Our technology enables your organization to mitigate risks in AI safety, security, and compliance while maximizing AI performance. Citadel AI’s technology is built from our team’s first-hand experience deploying high-risk AI systems at world-leading companies such as Google, Waymo, Toyota, and more.