OpenClaw 2026.3.2: PDF tool, safer defaults, stronger security

The newest OpenClaw release (v2026.3.2) focuses on one theme: making agent workflows more trustworthy for daily use. If you’ve been waiting to connect OpenClaw to real accounts or to run it in messaging-first setups, this update brings three practical upgrades: a production-ready secrets workflow, a first-class PDF tool, and safer defaults backed by a long list of security fixes.

What’s new in OpenClaw 2026.3.2

1) A first-class PDF tool for agents

OpenClaw now ships a dedicated pdf tool with native support for Anthropic and Google models (plus fallback extraction), so agents can read and reason over PDFs without awkward copy/paste steps. This makes common tasks like summarizing contracts, turning research reports into action items, or extracting structured fields from forms much easier to automate.

2) Secrets/SecretRef expansion (64 credential targets)

The secrets system expands SecretRef coverage across 64 credential surfaces and changes the failure behavior: unresolved references now fail fast on active surfaces, instead of silently breaking in the middle of a run. That’s a big quality-of-life win for operators who manage multiple providers, plugins, and channels.

3) Safer defaults + better config validation

New installs default the tool profile to messaging (rather than broad system access), and ACP dispatch is enabled by default, reducing the number of “gotchas” in first-time setups. There’s also a new CLI command, openclaw config validate (with --json), to lint configs before you start a gateway.

Why this matters for real workflows

If you’re building “AI staff” workflows (research, document triage, inbox routing, or multi-channel publishing), this release tightens up three points that usually cause pain: credential hygiene, document ingestion, and safe-by-default permission boundaries. The security hardening in the changelog also indicates the project is taking real-world threat models seriously.

Get the details

Read the full release notes on GitHub: OpenClaw v2026.3.2. If you’re new to this ecosystem, you can also explore more agent workflow ideas in our AI X Society hub: aixsociety.com.