That said, Hugging Face is an AI infrastructure company with the compute power readily available to host such large models. The big open-weight models need large amounts of VRAM to run and not many organizations have datacenters with enterprise-grade GPU clusters. In that case they will need to rely on neoclouds or services such as Amazon Bedrock or Microsoft’s Azure AI Foundry, after analyzing the jurisdiction, data retention, and privacy policies of these services, because simply using these open-weight models through their official APIs would mean sharing data with the Chinese labs that created them.
A multi-model architecture also needs strong identity controls, monitoring, target scoping, restricted tool permissions, network containment, evidence-handling rules, and human approval for important actions. The objective is resilience when a model is too restrictive for a particular task, not providing AI agents with autonomous access to every system in the environment.
“Organizations should have vetted AI models they can operate inside their own trust boundary before an incident happens,” Shah says. “That said, better models alone aren’t enough. AI for cybersecurity is still maturing, and organizations shouldn’t blindly trust autonomous systems.”
