Microsoft warns about risks in open-weight AI deployment
Microsoft backed open-weight AI models as key to US competitiveness, while warning developers lose control of model weights once deployed.
It said software guardrails can be stripped from locally modified open models, raising risks when models run agentic workflows tied to enterprise systems.
It urged hardware-anchored execution controls at “T=0” to block unauthorized execution threads at the silicon and PCIe level.
MSFT•
