Unified Policy Management for Diverse AI Stacks
Developers can now control how artificial intelligence agents interact with their systems using a new open-source project called Conduct. The tool acts as a runtime firewall for large language models and Model Context Protocol tools. It ensures that every AI action follows a single, unified policy. This approach applies to all team members and every individual agent session. The goal is to prevent unexpected behavior before it impacts production environments.
Breaking news
Eufy Unveils Local AI Home Security Ecosystem at IFA
The Rapid Evolution of Data Center Security in the AI Era
The High-Voltage Risks Facing Modern AI Data Centers
Apple’s New CEO Renames Lake Ontario To Lake America In Maps AppThe platform addresses a growing security gap in modern software development. As teams integrate more autonomous AI agents, managing their permissions becomes complex. Conduct provides a centralized layer of control. It sits between the application logic and the external tools the AI uses. This setup allows engineers to define strict rules once. Those rules then apply consistently across different programming languages and frameworks. The system reduces the risk of an agent executing dangerous commands or accessing sensitive data without approval.
The core feature of Conduct is its ability to enforce one policy across multiple surfaces. Traditional security tools often require separate configurations for each service. Conduct eliminates this fragmentation. It supports both direct large language model calls and interactions through the Model Context Protocol. The Model Context Protocol is a standard for connecting AI applications to external tools. By covering both areas, Conduct creates a comprehensive security boundary. Developers do not need to maintain separate rule sets for chatbots and tool-using agents. The repository includes two main product surfaces. These work together to monitor and validate every request. If an AI agent attempts an action that violates the defined policy, the system blocks it immediately. This prevents cascading errors or security breaches.
Why Do Teams Need Runtime Guardrails?
The project compares itself to existing runtime firewalls like Straiker and Lakera. However, Conduct focuses specifically on the interaction layer between AI models and external tools. While other tools might scan inputs or outputs, Conduct governs the execution of tools themselves. For example, if an agent tries to run a shell command, Conduct checks it against the policy first. This adds a critical safety check in the workflow. The open-source nature of the project allows teams to inspect the code and customize the logic. Companies can adapt the guardrails to fit their specific compliance needs without relying on a vendor lock-in.
AI agents are becoming more capable and autonomous. They can write code, execute scripts, and query databases. This power brings significant risks. A misconfigured agent could delete files or expose credentials. Conduct mitigates these risks by acting as a gatekeeper. The tool logs every decision made by the policy engine. This audit trail helps developers understand why a specific action was allowed or denied. It also simplifies debugging when an agent behaves unexpectedly. By centralizing governance, teams can scale their AI usage safely. New agents can be deployed quickly because they inherit the existing policy framework. This reduces the onboarding time for new AI components within an organization.
The release of Conduct highlights a shift in AI engineering priorities. Security and governance are moving from afterthoughts to core requirements. As more companies deploy agents in production, the need for standardized controls grows. Conduct offers a practical solution for this emerging market. It bridges the gap between flexible AI capabilities and rigid security requirements. The project aims to make safe AI deployment accessible to smaller teams as well. By providing a robust, open-source foundation, it lowers the barrier to entry for secure AI integration.
Frequently Asked Questions
Does Conduct support multiple programming languages? Yes, the tool operates at the runtime level. This means it can monitor and enforce policies regardless of the underlying language used to build the AI application.
How does Conduct handle the Model Context Protocol? Conduct integrates directly with the Model Context Protocol. It validates tool calls made through this standard, ensuring they comply with the defined organizational policies before execution.
Is the project fully open-source? Yes, Conduct is available as an open-source repository. Developers can clone the code, review the implementation, and contribute changes to improve the governance framework.
