OpenClaw 2.0 reorganizes the agent platform around its security seams

An agent that can act across your apps, browser, and services is only as safe as the seam between those layers. OpenClaw, the open-source platform that runs such agents, has reorganized its most recent release around that seam.

The 2026.8.1 release, internally branded "OpenClaw 2.0", touches installation, messaging, memory, skills, browser and native apps, plugins, and security. The project describes it as its largest update ever, built by 933 contributors including 569 first-time contributors and composed of more than 16,000 pull requests, a scale jump from the prior 106 releases, which shipped in 230 days, while this one took roughly seven weeks.

What is new in the security layer: better handling of credentials during agent execution, updated rules for how agents execute and how isolated they are from the host, new controls over which external tools and services agents can call, and changed plugin lifecycle management. Storage moved to SQLite, with a downgrade warning that requires restoring archived transcripts before rolling back.

Independent analyst Jaishiv Prakash, a Gartner director analyst, framed the work as structural rather than patch-level. "OpenClaw's latest release highlights that agent security cannot be addressed separately across runtime, memory and integrations," he said. "Because these components share identity, context and authority, a weakness in one layer can compromise the entire agent workflow."

The release lands against three 2026 CVEs in OpenClaw deployments (CVE-2026-32922, CVE-2026-25253, CVE-2026-26326) that drew outside scrutiny. It treats the layers together; it does not unwind the concentration.