According to OpenClaw on X, the OpenClaw v2026.3.8 release adds ACP provenance so agents can verify who is interacting with them, reducing spoofed identities in agent workflows (as reported by the OpenClaw release notes on GitHub). According to the GitHub release, the new openclaw backup feature enables rollback and state safety for rapid YOLO-style deploys, improving reliability for production AI agent operations. As reported by OpenClaw on X, duplicate Telegram events were eliminated, which stabilizes chat-based agent integrations and reduces redundant triggers. According to the GitHub release page, the update ships 12+ security fixes, signaling a hardening cycle that lowers operational risk for enterprises deploying AI agents. For builders, these improvements strengthen identity assurance in multi-agent systems, enhance disaster recovery, and cut integration noise—key for scaling agentic workflows in customer support, RPA, and chatbot orchestration.

According to @openclaw, the OpenClaw 2026.3.2 release adds Telegram live streaming, enables ACP subagents by default, introduces a native PDF tool, ships an openclaw config validate command, rebuilds the Zalo connector in pure JavaScript, and includes 100+ security and stability fixes (as reported on GitHub Releases and X). According to GitHub Releases, ACP subagents on by default can expand multi-agent orchestration out of the box, lowering integration effort for production agent workflows in customer support, content operations, and data extraction. As reported by @openclaw, Telegram live streaming enables real‑time conversational interfaces for AI agents, unlocking use cases like live commerce assistance and moderated community Q&A. According to GitHub Releases, the native PDF tool streamlines document parsing within the agent runtime, reducing reliance on external parsers and improving latency and privacy for enterprise document automation. According to GitHub Releases, the new openclaw config validate command improves DevOps reliability by catching misconfigurations in CI before deployment. As reported by @openclaw, the rebuilt Zalo integration in pure JS simplifies deployment across Node environments for Southeast Asia messaging automation. According to GitHub Releases, over 100 security and stability fixes strengthen agent reliability and harden interfaces, which is critical for regulated deployments and scaling multi-channel AI assistants.

According to @openclaw on X, the OpenClaw 2026.2.26 release adds External Secrets Management via openclaw secrets, enabling centralized rotation and least‑privilege access for agent credentials across environments, which reduces secret sprawl and audit risk for AI ops teams (source: OpenClaw on X). According to the same post, ACP thread‑bound agents gain a first‑class runtime that pins agent execution to conversation threads, improving determinism, state isolation, and reproducibility for multi‑turn LLM workflows in production (source: OpenClaw on X). As reported by @openclaw, Codex now defaults to WebSocket‑first transport, cutting request overhead and supporting streaming token delivery and bi‑directional events for lower latency agent actions, which benefits high‑throughput inference gateways (source: OpenClaw on X). According to @openclaw, the update includes an agent routing CLI with bind and unbind to dynamically attach agents to channels or tenants, unlocking faster A B testing and blue green rollouts of AI assistants (source: OpenClaw on X). As reported by @openclaw, Android app improvements enhance on‑device agent UX for mobile copilots, and 11 security hardening fixes address runtime and transport surfaces, strengthening compliance posture for regulated AI deployments (source: OpenClaw on X).

According to Andrew Ng, a new course on the Agent Communication Protocol (ACP) enables developers to build AI agents that can communicate and collaborate across different frameworks. The course, developed in partnership with IBM Research's BeeAI and taught by AI Research Engineer Sandi Besen, focuses on practical skills for implementing ACP in real-world scenarios. This training addresses a growing need for interoperability in multi-agent systems and presents significant business opportunities for organizations looking to integrate diverse AI technologies seamlessly. Source: Andrew Ng on Twitter.