StepSecurity
Malicious npm Releases Turn AI Memory Plugin into Credential Trap
StepSecurity reports that malicious releases of the npm package `@memtensor/memos-cloud-openclaw-plugin` turned a legitimate AI memory integration into a potential credential-harvesting entry point. Versions `0.1.21`, `0.1.23` and `0.1.25` contained a hidden `.sckit` executable, while the examined `0.1.22` and `0.1.24` artefacts were clean.
The malicious launcher was called during gateway startup and memory recall, ran a detached child process, inherited the host environment and, during recall, passed the user’s prompt text through `SCKIT_EVENT_TEXT`. Static analysis of six platform-specific Go binaries found patterns associated with cloud, source-control, npm, PyPI, AI-service and other credentials, alongside a configured home-directory inventory root and external fronts. The investigation did not execute the binaries and therefore did not confirm successful credential theft or data exfiltration.
The report also analyses the supplied `MemoryOS 2.0.34` PyPI source tree. A normal `memos` import can reach a logging-initialisation hook that attempts to launch a bundled `.sckit` binary, although execution on any particular host was not established. The package includes conditional CI helpers and native strings suggesting credential collection and propagation, but the custom build backend’s registration function was not called in the supplied source, and activation of the publishing path was not observed.
Organisations should remove or quarantine affected npm releases, treat `MemoryOS 2.0.34` as requiring investigation, preserve evidence, search for the listed files, processes and domains, and revoke credentials accessible to affected processes—particularly package-publishing, source-control and cloud credentials. They should also review prompts, package releases, workflow changes and repository activity for possible downstream compromise.