OpenAI Adds Sandboxing and In-Distribution Harness to Agents SDK for Enterprise Deployment
OpenAI has updated its Agents SDK with sandboxing capabilities that allow AI agents to operate in controlled environments, plus an in-distribution harness for frontier model deployment. The features launch initially in Python, with TypeScript support planned.
OpenAI Adds Sandboxing and In-Distribution Harness to Agents SDK for Enterprise Deployment
OpenAI has updated its Agents software development toolkit (SDK) with sandboxing capabilities and an in-distribution harness designed to help enterprises build and deploy AI agents more safely.
The SDK's new sandboxing feature allows agents to operate in controlled computer environments, accessing files and code only for specific operations while protecting overall system integrity. According to Karan Sharma from OpenAI's product team, the update makes the SDK "compatible with all of these sandbox providers."
The in-distribution harness enables agents to work with files and approved tools within a workspace when running on frontier models—the industry term for the most advanced, general-purpose AI models available. In agent development, a "harness" refers to all components of an agent system besides the underlying model itself.
"This launch, at its core, is about taking our existing agents SDK and making it so it's compatible with all of these sandbox providers," Sharma told TechCrunch. The goal is to allow users "to go build these long-horizon agents using our harness and with whatever infrastructure they have."
Long-horizon tasks refer to complex, multi-step work that requires sustained agent operation over extended periods.
Technical Details and Availability
The new capabilities launch initially in Python, with TypeScript support planned for a later release. OpenAI said it's working to add additional agent features including code mode and subagents to both Python and TypeScript.
The updated Agents SDK is available to all customers via the API using standard pricing. Pricing per specific operation was not disclosed.
What This Means
The sandboxing addition addresses a critical enterprise concern: agent unpredictability. By allowing agents to operate in isolated environments, companies can test and deploy autonomous systems without risking broader infrastructure. The in-distribution harness standardizes how agents interact with frontier models, potentially accelerating enterprise adoption by reducing custom integration work. This update positions OpenAI directly against Anthropic in the enterprise agent tooling race, though neither company has disclosed adoption metrics for their respective SDKs.
Related Articles
iOS 27 Code Reveals Hidden API to Replace Siri AI with Claude or ChatGPT
Developer PDFU discovered private hooks in iOS 27 and macOS 27 code that let apps register as Siri Extensions, potentially allowing users to swap Siri AI's backend for Claude, ChatGPT, or other third-party models. The feature is present in code but not currently enabled.
OpenAI's GPT-6 Astra Beats Claude Fable 5.1 Nearly 3-to-1 in Autonomous Business Benchmark, Tops Drone Navigation Tests
Independent testing lab Andon Labs found OpenAI's GPT-6 Astra nearly triples Claude Fable 5.1's performance running a simulated vending machine business, averaging $15,515 versus $5,422. Astra also became the first model to beat human-AI baseline performance across all five Drone-Bench subtasks, including autonomous person-tracking via drone.
GPT-6 Astra Beats Ai2's MolmoAct2 on New Robotics Benchmark, Researcher Calls It a 'Step Change'
A new robotics benchmark called StationeryBench shows OpenAI's GPT-6 Astra completing 7 of 100 desk-object manipulation tasks versus zero for Ai2's MolmoAct2, with a median progress score of 46 against 12. Cornell/DeepMind researcher Yoav Artzi calls the result a 'step change in spatial reasoning.'
Perplexity Says It Runs End-to-End Engineering Systems on OpenAI's GPT-6 Astra
Perplexity says it has shifted core engineering workflows, including code changes and production monitoring, onto OpenAI's GPT-6 Astra model. The claim comes from an OpenAI-published case study with no independent benchmark data released.
Comments
Loading...