OpenClaw Update March 2026: Backup CLI, ACP Provenance, and 12+ Security Patches

v2026.3.8 is already running on your ClawHosters instance. No action, no downtime, no terminal required.

OpenClaw v2026.3.8 shipped on March 8 with contributions from 43 developers. It's a reliability and hardening release. Not flashy, but the kind that matters most in production.

Here's what changed. It's mostly hardening -- exactly what production infrastructure needs.

OpenClaw Backup CLI

The new openclaw backup create and openclaw backup verify commands let self-hosters archive their agent state locally. You can grab everything or use --only-config for a config-only snapshot.

For ClawHosters customers, this probably isn't relevant day-to-day. Your backups are handled automatically -- why ClawHosters handles this for you is worth understanding if you're deciding between running your own stack versus managed hosting. But if you're curious about what's available under the hood, the official release notes have the full command reference.

ACP Provenance

One of the standout changes in this March 2026 update is ACP provenance -- think of it as caller ID for agent-to-agent communication. When another agent or system talks to yours, ACP provenance metadata now lets your agent verify who initiated the conversation.

The feature is opt-in with three modes: --provenance off, meta, or meta+receipt. Most individual users won't touch this directly. But for anyone running multi-agent setups, Blockchain.news describes it as reducing spoofed identities in agent workflows. That matters more in practice than the name suggests: spoofed agent identities are how multi-agent pipelines get poisoned with bad data or unauthorized instructions.

12+ OpenClaw Security Patches

This OpenClaw update March 2026 bundles over a dozen security fixes spanning SSRF protections, script execution sandboxing, and download-tool pinning. As The Droid Guy summarizes, the patches also cover plugin route hardening and archive extraction safety. It follows the v2026.2.6 safety scanner release, which introduced the built-in scanner for monitoring agent behavior.

The number reflects a planned hardening cycle, not a response to active exploits. If you're self-hosting, update now. On ClawHosters, these patches went live the day they shipped. If you want to understand the full threat model, our complete OpenClaw security hardening guide covers what to lock down beyond updates.

Other Fixes

• Telegram duplicate replies: If your agent ever replied twice to the same message, that bug is gone. Deduplication now tracks per agent instead of per session key.

• Brave Search LLM context mode: The new llm-context mode returns pre-formatted snippets with source metadata. Your agent's web search results get more useful, more reliably.

• Gateway restart reliability: The gateway now exits non-zero when draining takes too long, so service supervisors actually restart it. Mostly relevant for self-hosters, but ClawCloud's writeup notes this fix alone is worth updating for.

This openclaw update march 2026 reflects the project's reliability-first approach. If you'd rather not manage updates yourself, all ClawHosters plans handle this automatically.