San Francisco — February 23, 2026 — NEAR AI announced the launch of IronClaw, an open-source AI agent runtime designed to support always-on autonomous agents with enhanced security and privacy protections. The product was introduced on stage at NEARCON 2026.
IronClaw builds on the earlier OpenClaw concept and incorporates additional cryptographic safeguards. Developed in Rust and deployed within an encrypted Trusted Execution Environment (TEE) on NEAR AI Cloud, the runtime is designed to allow AI agents to access tools and act on behalf of users while limiting exposure of sensitive data and credentials.
According to NEAR AI, IronClaw is intended to provide a controlled execution environment for AI agents that require persistent memory, tool access, and task automation. The system is structured to give users defined control over data access and asset permissions. A free Starter tier includes one hosted AI agent instance running within NEAR’s infrastructure, with paid tiers available for expanded deployment and scaling.
Security architecture is central to the runtime’s design. IronClaw uses a defense-in-depth model that applies multiple layers of protection to reduce misuse and unintended data exposure. Third-party or untrusted tools operate within isolated sandbox environments and can only access explicitly authorized resources. Network communications are restricted to approved endpoints, and sensitive credentials are injected at runtime rather than stored in exposed formats.
Illia Polosukhin, Co-Founder of NEAR Protocol and Founder of NEAR AI, said the goal was to create a trusted infrastructure layer for AI agents. He described IronClaw as a security-focused framework that combines confidential inference capabilities with built-in safeguards, extending NEAR’s broader approach to infrastructure design into the AI layer.
IronClaw also incorporates monitoring mechanisms intended to mitigate risks such as data leakage, misuse of agent permissions, and prompt injection attacks. The runtime enforces resource usage limits and continuously evaluates agent activity within its controlled environment. User data is stored locally in PostgreSQL databases, with personal information encrypted using AES-256-GCM standards. The platform maintains detailed audit logs of tool activity while, according to NEAR AI, collecting no telemetry or analytics data that would transfer information outside the user’s environment.
George Zeng, Chief Product Officer and General Manager of NEAR AI, stated that as AI agents become integrated into operational workflows, security, compliance, and data governance concerns have grown. He said IronClaw is intended to provide developers and enterprises with infrastructure to deploy agent-based systems while maintaining oversight and control.
IronClaw is now available for deployment, and the source code has been published on NEAR AI’s GitHub repository.
