Loading vulnerability…
OpenClaw versions prior to 2026.2.14 contain a flaw in the gateway’s handling of node.invoke parameters where internal approval/control fields are not properly sanitized. An authenticated client can inject these internal approval fields to bypass exec approval gating for system.run commands, resulting in the ability to trigger arbitrary command execution on connected node hosts (e.g., developer workstations and CI runners) through the gateway.
Mallory correlates every CVE against your assets, your vendors, and active adversary campaigns. Know which vulnerabilities matter for you, not just which ones are loud.
What it means. What to do now. Patch path, mitigations, and the assume-compromise checklist.
What an attacker gets, and what they’ve been doing with it.
If you can’t patch tonight, do this now.
Patch, then assume compromise.
1 valid exploit after Mallory filtered fakes, detection scripts, and README-only repos.
This repository is a small, single-purpose Python exploit for CVE-2026-28466 affecting OpenClaw versions earlier than 2026.2.14. The repository contains only two files: the main exploit script (CVE-2026-28466.py) and a minimal README referencing the affected versions and GitHub advisory. The exploit is a network-based WebSocket RCE tool. It accepts a target gateway address and an arbitrary command from the operator, then connects to the target at ws://<host>:<port>/ws using a spoofed Origin header. After receiving a connect.challenge event, it sends a connect request that impersonates the OpenClaw web control client (client id openclaw-control-ui) and requests operator/admin scopes. Authentication relies on a hardcoded token embedded in the script. Once authenticated, it issues a node.invoke request against a hardcoded node ID, invoking the system.run command with the attacker-supplied command and approved=true. The script then prints stdout, stderr, and exitCode from the response. Structurally, the code is straightforward and operational rather than a mere proof-of-concept: it implements the full WebSocket message flow, handles authentication, generates an idempotency key, submits the remote execution request, and parses returned output. However, it is not highly flexible because critical values such as the token and node ID are hardcoded. There is no persistence, staging, lateral movement, or post-exploitation automation; the core capability is authenticated remote command execution against an OpenClaw node through its WebSocket API.
Products and vendors Mallory has correlated with this vulnerability. Open in Mallory to drill down to specific CPE configurations and version ranges.
Vendor-confirmed product mapping. Mallory continuously reconciles this list against your asset inventory.
10 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.
Query your assets running an affected version, and investigate the blast radius.
Every observed campaign linking this CVE to a named adversary.
Malware families riding this exploit, with evidence and IOCs.
YARA, Sigma, Snort, and vendor rules, auto-deployed to your SIEM.
Cross-references every affected SKU, including bundled OEM variants.
Community discussion across Reddit, Mastodon, and other social sources.