NFOX Lab, an artificial intelligence governance startup that emerged from stealth, launched a platform for managing owned large language models and announced plans to open-source Blockhouse, a security sandbox designed to restrict what AI workloads can access and do, SiliconANGLE reported.[1] That addresses a dilemma for companies that want AI capabilities but hesitate to send sensitive information to external providers.[1] Blockhouse removes privileges by default and enforces access policies at the Linux kernel level, operating beneath the application layer to limit what models can do even if compromised.[1]
The company's approach borrows from zero-trust security principles, and Gordon Allott, nFOX's founder and chief executive, said, "Every model has the potential to be hostile."[1] Rather than relying on instructions to guide behavior, Blockhouse removes capabilities that could enable unauthorized actions, according to Allott, and uses a monitored channel called Porthole for outbound requests.[1] Runtime credentials and audit controls remain outside the model process, and administrators can restrict file access and approve destinations through whitelists.[1] SiliconANGLE reported that approved model files are sealed with SHA-256 hashes before serving, and that built-in checks halt serving if those files change.[1]
Allott acknowledged that human error remains a risk, such as mistakenly authorizing a malicious destination.[1] The sandbox supports the vLLM, SGLang and llama.cpp inference engines, and the accompanying nFOX Platform offers a console for evaluating, deploying and monitoring models across cloud, rented GPU hosts and on-premises Linux servers.[1] The nFOX Platform is available now, while Blockhouse is in beta for qualified enterprises, SiliconANGLE reported.[1] Allott said nFOX plans to release Blockhouse under the Apache 2.0 license, while charging an annual license for its fleet management platform, and intends to incorporate NVIDIA’s OpenShell, a similar agent security technology.[1]