OpenAI updates Agents SDK with enterprise security and multi-agent features

OpenAI has updated its Agents SDK, a toolkit for developing autonomous AI agents, with a focus on security and enterprise-scale capabilities. The company expanded the framework in response to growing demand from large organizations that want to build complex agent systems with predictable, controllable behavior.

The Agents SDK, first introduced in early 2025, was originally aimed at developers building agent prototypes. The new update shifts the emphasis toward production-ready solutions: tracing of agent workflow execution has been improved, and guardrails, the mechanisms that validate incoming and outgoing data, have been expanded. Enterprise customers can now control more precisely what context an agent receives and what it returns to the user, preventing unwanted or unsafe responses.

One of the key additions is better support for multi-agent orchestration. The SDK now has more flexible handoff mechanisms, which pass control between agents within a single workflow. This matters especially for complex enterprise scenarios in which a task is processed in sequence by several specialized agents: one searches for information, another analyzes it, a third produces the final answer. The new version makes such chains more reliable and more transparent for developers.

OpenAI also reported improvements to its observability tools. Integration with tracing systems lets teams see in real time which steps an agent took, which tools it called and at what stage errors occurred. This is critical for auditing in regulated industries such as banking, healthcare and the legal sector, where every automated decision must be explainable and documented.

Big business interest in agentic AI systems continues to grow. According to analysts, the number of enterprise pilots with AI agents more than doubled over the past year. Companies are looking for ways to automate multi-step workflows, from handling customer requests to managing IT tickets and analyzing documents. OpenAI positions the updated SDK as a tool that lowers the barrier to entry for enterprise developers, who until now have had to build agent infrastructure from scratch.

Competition among agent frameworks is intensifying. Anthropic is developing the Claude Agent SDK with an emphasis on constitutional safety. Google offers the Agent Development Kit with native integration into Vertex AI. Microsoft promotes AutoGen and Semantic Kernel, while independent frameworks such as LangChain retain a share among open-source developers. OpenAI's update is part of the race for the trust of the enterprise sector, where security and compliance requirements are significantly stricter than in consumer applications.

For the market as a whole, this is a signal: agent systems are ceasing to be an experimental technology and becoming a production standard. Major vendors are investing not just in launching agents but in tools for managing them over the long term, with versioning, monitoring and compliance with corporate security policies.