OPENCLAW continues to reveal security failings even as its creator moves toward a new OpenAI-backed foundation, with OpenClaw’s February 14, 2026 announcement that Peter Steinberger has joined OpenAI marking the latest transition.
According to SecurityWeek, CVE-2026-25157 was fixed on 25 January in version 2026.1.25, and a one-click RCE vulnerability (CVE-2026-25253) disclosed on 1 February was patched on 29 January in 2026.1.29, though the Docker sandbox bypass (CVE-2026-24763) required a further fix in 2026.1.30; additional fixes addressed CVEs 2026-25593 and 2026-25475, with version 2026.2.17 cited as the latest and no known unfixed CVEs remaining.
The piece notes a large legacy population still using older builds, making those CVEs exploitable, while also highlighting widespread AI agent misconfiguration risks. Alex Polyakov has launched SecureClaw, an open-source tool to audit OpenClaw deployments, described as the first open-source security solution purpose-built for OpenClaw, and it runs 55 automated checks mapped to OWASP, MITRE ATLAS, and CoSAI Agentic AI Security.
SecurityWeek also covers the OpenClaw supply chain attack ClawHavoc and incidents tied to Moltbook-style exposure and credential harvesting, underscoring that even as newer versions ship, many deployments remain at risk.