# OpenAI Upgrades Agents SDK with Native Sandbox Execution

OpenAI announced a major update to its Agents SDK, introducing native sandbox execution and a model-native harness designed to help developers build more secure and capable AI agents.

The upgrade addresses two critical needs in AI agent development: security and persistence. The native sandbox execution provides an isolated environment where agents can run code and interact with files without risking the broader system. This is crucial for developers who need their AI agents to perform complex tasks autonomously while maintaining security standards.

The model-native harness enables agents to run for extended periods and work seamlessly across multiple files and tools. This means developers can now build agents that handle more sophisticated, multi-step workflows without constant human intervention.

These improvements come as the AI industry shifts toward "agentic" systems—AI that can plan, execute tasks, and use tools independently rather than simply responding to prompts. OpenAI's