Agent Factory: Formulating a plan for safe and reliable AI agents
By Yina Arenas, Corporate Vice President of Azure AI Foundry
Azure AI Foundry introduces a structured process that combines security, safety, and governance for enterprises to establish trust in their agents.
Trust is increasingly becoming the primary obstacle for enterprise AI. While observability focuses on visibility, security is about steering. As agents transition from innovative prototypes to fundamental business systems, companies are grappling with a challenging query: how can we ensure the safety, security, and manageability of agents as they expand?
The solution does not lie in quick fixes here and there. Rather, it involves a blueprint. An approach that prioritizes trust by integrating elements like identity management, guidelines, assessments, adversarial testing, data security, surveillance, and regulatory compliance in layers.
Across various sectors, similar concerns are voiced: Chief Information Security Officers (CISOs) are troubled by agent proliferation and ambiguous ownership. Security teams require guidelines that align with their existing processes. Developers seek to embed safety from the project’s inception instead of adding it as an afterthought.
These challenges are leading to the emergence of the shift-left trend. Responsibilities related to security, safety, and governance are being shifted earlier into the development workflow. Teams cannot afford to wait until deployment to secure agents; they need inherent safeguards, assessments, and policy integration right from the beginning.
Issues such as data breaches, prompt tampering, and regulatory ambiguity continue to hinder AI adoption. For companies, trust has now become a critical factor in determining whether agents progress from trial phases to full-fledged production.
In terms of enterprise adoption, five key attributes stand out: Distinct identity markers, Data safeguarding integrated from the outset, Inbuilt controls, Evaluation against potential threats, Ongoing supervision.
While these attributes do not ensure absolute protection, they are crucial for constructing trustworthy agents that meet enterprise standards. Incorporating these aspects into our products reflects Microsoft’s commitment to ethical AI practices.
Azure AI Foundry unites security, safety, and governance features in a structured process that enterprises can adopt to instill trust in their agents.
Soon, every agent generated within Foundry will be assigned a unique Entra Agent ID for organizations to track all active agents across their network and minimize unauthorized ones.
Foundry boasts comprehensive built-in agent controls—a pioneering feature in the industry. It is the sole AI platform equipped with a cross-prompt injection classifier that scrutinizes not only prompt documents but also tool responses, email triggers, and other dubious sources to identify and neutralize harmful commands.
Assessments establish a feedback loop throughout the lifecycle. Teams can conduct risk assessments, groundedness evaluations, and protected material scans before deployment as well as during operational phases.
Through the standard agent configuration in Azure AI Foundry Agent Service, companies can utilize their own Azure resources such as file storage, search capabilities, and conversation history archives.
Foundry Agent Service supports network isolation via custom virtual networks and subnet delegation for enhanced privacy protection.
Microsoft Purview extends data security measures and compliance protocols to AI operations.
Foundry presents alerts and suggestions from Microsoft Defender directly within the agent environment.
Foundry collaborates with governance partners like Credo AI and Saidot.
In terms of enterprise adoption best practices include: Commence with identity verification procedures; Embed controls from the start; Continuously assess performance; Safeguard sensitive data; Employ enterprise tools for monitoring; Align governance with regulations.
Enterprises are already formulating security strategies using Azure AI Foundry.
Begin your journey with Azure AI Foundry. Join us at Microsoft Secure on September 30th for insights into our latest capabilities.
Did you overlook these entries in the Agent Factory series?
The future of AI commences here. Visualize your next groundbreaking AI application with cutting-edge technologies. Kickstart your project with Azure.