Archive — US — techcrunch
OpenAI updates its Agents SDK to help enterprises build safer, more capableagents
OpenAI has updated its Agents SDK, adding features such as sandboxing and an in-distribution harness to help enterprises build safer, more effective AI agents. These tools will allow agents to operate in controlled environments and handle complex, long-horizon tasks while protecting system integrity. The updates initially support Python, with TypeScript on the way.
Key points
- OpenAI updates the Agents SDK to enhance the safety and capability of AI agents.
- New features include sandboxing, allowing agents to work in controlled environments.
- Sandboxing prevents risky, unsupervised operations by isolating agents in specific workspaces.
- The in-distribution harness enables agents to use files and approved tools efficiently.
- OpenAI's goal is to empower developers to create 'long-horizon' agents for complex tasks.
- Initial support for these capabilities is in Python, with TypeScript planned for future release.
Why it matters
- The trend towards responsible AI development aligns with increasing regulatory scrutiny and public concern over AI safety.
- Similar initiatives from other tech companies, like Anthropic, highlight a competitive focus on agent development.
- The push for safer AI reflects broader IT trends in 2025 towards enhanced cybersecurity and more robust machine learning models.
In summary, OpenAI's updates to the Agents SDK mark a strategic step towards safer, more capable AI systems, addressing both enterprise needs and the rising importance of responsible AI development.