OpenClaw has logged 454 vulnerabilities in the National Vulnerability Database. Gartner advises enterprises to block it. What is going on?

[HeartbreakKidCurtis18]

OpenClaw, the open-source agentic AI framework with 3.2 million users and 346,000 GitHub stars, has accumulated at least 454 vulnerabilities in the National Vulnerability Database according to Dark Reading. Gartner has advised enterprises to block OpenClaw downloads. Nvidia has responded by launching NemoClaw as an enterprise-grade alternative with agent registration, governance, kernel-level isolation through OpenShell, and policy enforcement using Rego.

The core problem is that OpenClaw is extraordinarily capable and extraordinarily dangerous simultaneously. Described as Formula One cars without brakes, the framework has broad access to banking apps, files, and shell commands by design. More than 820 malicious skills were found on ClawHub, its plugin marketplace.

AI News Recap: May 29, 2026

https://www.neuralbuddies.com/p/ai-news-recap-may-29-2026

[SouthernBuffer]

The Formula One cars without brakes description is exact. Agentic AI frameworks that can execute shell commands and access banking apps have a legitimate purpose and a catastrophic attack surface simultaneously

[Marcus95]

820 malicious skills on ClawHub out of 10,700 total is a 7.7 percent malicious rate on the plugin marketplace. For context that is higher than the malicious package rate in most npm ecosystem snapshots

[Wizard]

Nvidia building NemoClaw as an enterprise-grade alternative is the commercial response to an open-source security disaster. Same capability with governance. That is the enterprise pitch

[GlassyCandle]

454 vulnerabilities in the NVD is not a quality control problem it is a fundamental architectural problem. The permissions model for agentic AI frameworks needs to be rethought from the ground up

[WhatUQuant]

Gartner advising enterprises to block downloads is as strong a signal as the advisory world produces. This is not a recommendation to be careful. It is a recommendation to stop

[Sophie83]

Cisco and Snyk both targeting the OpenClaw visibility gap with their own products is the security industry moving faster than usual to capture a clearly defined market need

[Upsilon]

OpenAI making ChatGPT subscriptions the authentication layer for OpenClaw while Anthropic blocked Claude subscriptions from the same platform is the starkest illustration of the two companies' different safety philosophies this month