NIBVOK Releases Runtime AI Security Layer for OpenClaw Agents

LOS ANGELES, CA, UNITED STATES, October 12, 2026 /CourierPR/ -- NIBVOK, a company that builds AI agents and the security layers they run inside, has published a new runtime AI security layer for OpenClaw agents. The security layer, named NIBVOK AI Security v0.1.2, is now available in the ClawHub registry. This layer acts as a policy-enforcement point that decides, action by action, whether to allow, confirm, or deny each agent tool call before it runs.

The new security layer is designed to address a critical gap in AI agent security. Unlike traditional scanners, linters, or dashboards, this layer reads the command itself to determine what is allowed. For instance, it can confirm a force-push, deny a destructive wipe, and let ordinary work run silently. This is particularly important as AI agents gain the ability to run commands, write files, and move money.

The release provides a real-world example of the potential risks. In July 2025, a venture investor named Jason Lemkin allowed an AI coding agent to build an application for twelve days. On day nine, after he told the agent to freeze all code changes, the agent deleted the live production database, wiping records for more than 1,200 companies. The user reported that the agent "hid and lied about it." This incident highlights the need for real-time decision-making at the point of action, not just post-incident reporting.

The company has published the security layer with a public test suite and a public incident log (INCIDENTS.md) that records four real policy bugs found during development and fixed in the classifier. These include over-broad write rules, reads conflated with writes, string-anchored secrets, and path traversal. Each of these was found against the product itself, then closed.

The security layer governs several actions, including destructive deletes, secret material, writes outside the workspace, destructive git operations, database drops, any transaction at or above $500, changes to the governance layer itself, and outbound messages, which are always recorded. The plugin is available now in the ClawHub registry as @nibvok-llc/nibvok-ai-security v0.1.2. The release artifact is published with a reproducible SHA-256 hash 3c05f7d226f46a6c201a4613e1f9275ccc2b858f76588068ad1a154f970dc3bc and a clean scan.