Malicious code in claw-subagent-service (npm)
Exploitation and risk
Source statusCISA KEV lists vulnerabilities known to be exploited. FIRST EPSS estimates the likelihood of exploitation in the next 30 days. An unavailable score does not mean zero risk.
Which CVEs have scores here?
This view includes predictions for CVE identifiers explicitly listed in public NuGet advisories. Microsoft release pages have their own prediction coverage. Check source status for the scope and latest checks.
No exploitation or prediction data is available here for these CVEs. This does not mean they are safe or unlikely to be exploited.
osv · MAL-2026-3757
Published--- _-= Per source details. Do not edit below this line.=-_ ## Source: amazon-inspector (b2ccba152d6841731431c91157874c72b5f9778fdf88b634a45ab5d9da961307) On `npm install -g`, the package's `scripts/post-install.js` registers a privileged Windows service `claw-subagent-service` pointing at `service/daemon.js`, configured with `sc.exe failure... actions= restart/0/restart/0/restart/0` and `start= auto`, then immediately starts it — running as LocalSystem on Windows with no opt-in prompt. Once running, three behaviors stack into a remote-control surface against the installer: 1. Self-replacing update channel (`service/updater.js`): a 6-hour interval calls `npm view claw-subagent-service version --json` and, when a newer version is published, runs `npm install -g claw-subagent-service@<latest>` and restarts the worker. Any future tarball published under this name is fetched and executed under LocalSystem (Windows) / user (macOS) without consent and without honoring the operator's pinning. Linux is gated, Windows + macOS are not. 2. Vendor-controlled IM command channel (`service/worker.js`, `service/modules/rongyun-message-handler.js`, `service/modules/script-executor.js`, `service/rongcloud/openclaw-client.js`): the worker fetches a token from `https://newsradar.dreamdt.cn/im/api/claw/token/<nodeId>` and joins a RongCloud IM session (appKey `bmdehs6pbyyks`). `RongyunMessageHandler` dispatches inbound `COMMAND` / `DEVICE_CONTROL` / `CHAT_MESSAGE` messages to handlers that `spawn` `start.sh`/`stop.sh`/`restart.sh`/`status.sh`, run `openclaw doctor --fix`, and feed attacker-supplied prompts into the local AI agent (which can in turn invoke arbitrary tools). Whoever controls the vendor IM backend (or its appKey) has persistent privileged shell-class access to every installer machine. 3. Continuous data exfiltration (`service/modules/heartbeat-dashboard.js`, `service/modules/dashboard-collector.js`): every 30 seconds the worker reads `~/.openclaw/agents/*/sessions/*.jsonl`, `~/.openclaw/projects/projects.json`, `~/.openclaw/tasks/tasks.json`, and the host MAC address, and ships them as 6 RongCloud IM messages to the vendor backend. Includes session contents, model-provider metadata, token-usage events, and a stable host identifier. The README documents the product's purpose, but the combination — postinstall privileged-service persistence + 6-hour silent self-replacement + always-on remote-command IM channel + continuous session/host-id upload — is a vendor-operated remote-administration agent installed on the operator's machine via `npm install`. A compromise of the publisher account or the vendor IM backend yields immediate, unattended code execution on every installer host.
- Source published
- 14 May 2026 19:25:16 UTC
- Source modified
- 5 Aug 2026 17:21:45 UTC
- Saved here
- 24 Oct 2026 13:17:17 UTC
- Source terms
- OSV · upstream source license applies
Original source record · Cite this exact observation · Complete original publisher JSON
Source: osv · OSV · upstream source license applies. The publisher's wording is preserved. ChangeIntel groups the information and labels changes. The JSON record includes additional source fields.