Nemoclaw launches publicly
- Nemoclaw debuted as an AI tool claiming to find security flaws and suggest optimizations for models. (x.com) - The initial Poseidanai demo video showed live scans and had about 26 views on release. (x.com) - Early reaction has been mixed, with enterprise breakdowns and critiques of its three-layer security model. (x.com) (x.com)
NVIDIA publicly released NemoClaw, an open‑source security stack for OpenClaw agents, at GTC on March 16, 2026. (nvidianews.nvidia.com) NemoClaw installs the new NVIDIA OpenShell runtime and can route inference to NVIDIA Nemotron models with a single-command alpha preview. (docs.nvidia.com) The code and installer live in the NVIDIA/NemoClaw GitHub repository, which is marked alpha and shows active commits and community activity. (github.com) OpenClaw’s rapid adoption and a string of public security findings in early 2026 motivated vendors to add hardened runtimes for enterprise deployment. (zdnet.com) NemoClaw’s deployed architecture layers kernel-level sandboxing (Landlock, seccomp, net namespaces), out‑of‑process policy enforcement via OpenShell, and a privacy router that controls where inference runs. (particula.tech) Security researchers have given mixed assessments: a Cloud Security Alliance note described NemoClaw as a “practical middle ground,” while hands‑on reviews flagged governance gaps and remaining attack surface. (labs.cloudsecurityalliance.org) Independent audits and writeups from teams that ran the stack reported configuration traps, docs gaps and performance tradeoffs for local inference on GPUs. (ongil.ai) A short demo clip shared on X by creator Poseidanai showed live scans of the stack in action and, by social counts at release, attracted only a few dozen views. (x.com) NemoClaw remains alpha; NVIDIA’s docs and the project’s GitHub issues page show ongoing fixes and invite external reporting, so expect more patches and policy updates in coming weeks. (github.com)
