NEAR AI ships IronClaw 1.2 with multi-user agents and open tool connectivity

@NEARProtocol AI has released version 1.2 of @IronClawAI, its open-source secure agent platform, expanding the runtime with multi-user support, broader tool connectivity, and tighter defenses against prompt injection.

Open Tool Connectivity via MCP

The headline addition in version 1.2 is support for any MCP-ready tool, stretching from CRMs to internal APIs. IronClaw's Model Context Protocol (MCP) integration lets users connect to MCP servers for additional capabilities, with a plugin architecture that allows new WASM tools and channels to be dropped in without restarting the agent. That flexibility lets teams wire the agent into existing enterprise tooling without rebuilding their stack.

Agents can now operate inside Slack channels and Telegram groups directly, with every message processed under the sending user's own identity, permissions, and approval gates. IronClaw supports multi-channel access through Slack, Telegram, and a web gateway, alongside background routines, heartbeat-based monitoring, and parallel jobs.

Security Model: Untrusted History, Encrypted Enclaves

The update also sharpens how IronClaw handles potentially hostile content in shared rooms. Channel history is treated as untrusted content rather than instructions, and the platform guards against prompt injection when agents read public channels. IronClaw enforces strict limits on resource usage and applies advanced protections against prompt injection attacks.

Built in Rust and deployed inside an encrypted Trusted Execution Environment (TEE) on NEAR AI Cloud, IronClaw is designed to enable autonomous agents with built-in privacy safeguards and controlled access to user assets and data. Crucially, user secrets are kept outside the model itself. API keys, tokens, and passwords are stored in an encrypted vault, with IronClaw injecting them only where the user has permitted access. The AI never sees the raw values.

All data remains stored locally within a user's PostgreSQL database, encrypted using AES-256-GCM. IronClaw also maintains a comprehensive audit log of all tool activity while collecting no telemetry, ensuring data never leaves the user's local environment.

The broader security rationale is straightforward. AI agents are powerful, and that power without security is a liability. These systems can hold credentials, browse the web, and take actions on a user's behalf. Without guardrails, they become a threat to privacy, data, and trust. IronClaw's architecture is built around exactly that concern, keeping execution verifiable and secrets hardware-enforced.

Co-founded by Illia Polosukhin, NEAR Protocol has evolved from an Ethereum-like smart contract platform to prioritising AI integration and data privacy. Polosukhin is also a co-author of the landmark "Attention is All You Need" paper that pioneered Transformer models.

Sources:
Forbes: IronClaw Meets OpenClaw And AI Agent Security Gets Serious
AIThority: NEAR AI Launches IronClaw, A Secure Runtime for Always-On AI Agents
IronClaw Official Documentation