Azure AI Foundry releases Responses API and Computer Action Agents (CUA)

AI agents are reshaping all industries, driving automation, improving productivity, and enabling intelligent decision-making. Companies have widely used AI agents to process insurance claims, manage IT services, optimize supply chain logistics, and even assist doctors in analyzing medical records. The potential of intelligent technology is constantly expanding. Recently, Azure AI Foundry officially launched two major innovations:

• Responses API : A powerful API that helps AI applications efficiently obtain information, process data, and perform actions, making intelligent decision-making smoother.

• Computer Operation Agent (CUA) : A breakthrough AI model that can autonomously operate software interfaces, perform tasks, and automate workflows. 

These two innovations will help companies further unlock the potential of AI technology, making AI not just an assistant but also a virtual workforce for companies, thereby promoting large-scale automation, improving efficiency, and accelerating intelligent upgrades. 

In Azure AI Foundry, Responses API is the key to building "intelligent AI" and bringing stronger intelligent productivity to enterprises. It is not only the new cornerstone of the built-in tools of Azure OpenAI service (international version) , but also combines  the simplicity of Chat Completions API, the advanced capabilities of Assistants API and Azure AI Agent Service , making it easier for enterprises to empower AI technology into business processes. 

With the Responses API: 

✅ Only one call is needed to achieve seamless interaction with CUA, function execution, file search and other tools, and realize smarter task automation .  

✅Data retrieval, information processing, task execution , seamless integration of reality and multiple workflows. 

The Responses API provides a structured response format that allows AI agents to interact with multiple tools while maintaining contextual memory during the interaction. It supports: 

By integrating retrieval, reasoning, and execution operations into one API, the complexity of AI agent development is greatly reduced, allowing enterprises to easily build automated workflows without having to worry about the orchestration of multiple AI tools. 

This scalability is ideal for industries such as customer service, IT operations, finance, and supply chain management, where automation driven by AI technology can simplify workflows and improve efficiency. For greater flexibility and management capabilities, enterprises can also combine Azure AI Agent Service. Azure AI Agent Service provides richer tools and models, supports Semantic Kernel and AutoGen, and allows multiple agents to collaborate efficiently to cope with more complex business scenarios. 

Computer Operation Agent (CUA) is  a special AI model of Azure OpenAI service (international version) that has the ability to autonomously operate graphical user interfaces (GUIs) . Through natural language instructions, the computer operation agent can interact with applications and automatically perform multi-step operations, interpret visual elements, dynamically adapt and take further actions based on the screen content. 

✅Autonomous UI interaction : You can open applications, operate interfaces, and complete multi-page tasks without preset scripts. 

✅Dynamic adaptation : Intelligently identify UI changes, flexibly adjust execution steps, and reduce dependence on preset automation scripts.

✅Execute tasks across applications : You can operate Web and desktop applications at the same time without API adaptation, breaking system barriers.  

✅Natural language command interface : Users only need to describe the task in natural language, and CUA can automatically parse and perform the corresponding operations. 

In addition, we are also exploring the deep integration of CUA with Windows 365 and Azure Virtual Desktop (AVD) . In the future, enterprises can host CUA in the cloud to achieve large-scale AI automation while ensuring compliance and security. This is not only the evolution of intelligent bodies, but also a new chapter in AI productivity!  

As the autonomy of AI agents continues to increase, how to ensure their security, reliability, and compliance with user intent has become a core issue. As one of the first agent models that can directly operate the software environment, computer operation agents (CUA) not only improve automation capabilities, but also bring new challenges such as anti-abuse, avoidance of misoperation and adversarial risks. To this end, Microsoft and OpenAI have built a set of multiple security protection mechanisms covering models, local systems and enterprise deployment to ensure that AI applications are safe and controllable. 

• Model-level security : CUA has built-in security policies that deny malicious tasks, block unauthorized operations, and prevent abuse. 

• System-level monitoring : Microsoft provides enterprise-level content filtering and execution monitoring to detect and block illegal operations in real time. 
  

•  Confirmation of critical tasks : To reduce the risk of misoperation, CUA has designed a user confirmation mechanism to remind users before performing irreversible tasks and limit high-risk operations (such as financial transactions). 
  

• Enterprise compliance assurance : Based on the Microsoft Trusted AI framework, it provides real-time observability, logging, and compliance auditing to ensure that enterprise deployment and operation are transparent, compliant, and controllable. 
  

• Risk detection and enhancement : Microsoft combines automation and manual review mechanisms to monitor AI execution patterns and identify abnormal behaviors. Through internal testing, external audits, and real-world scenario testing, Microsoft continuously optimizes security policies to prevent prompt injection, adversarial attacks, and unauthorized access.

CUA is still in the process of continuous optimization, especially its reliability in non-browser environments still needs to be further improved. Therefore, for operations involving high sensitivity , we still recommend maintaining manual supervision . 

As AI agents continue to evolve, Microsoft will continue to strengthen transparency, security, and risk prevention, and combine Azure's enterprise compliance and governance tools to ensure that enterprises can deploy AI automation safely and compliantly at scale.