Microsoft, Nvidia push provenance checks as open-weight AI adoption accelerates

Summary

Microsoft and Nvidia urge enterprises to assess open-weight AI models throughout deployment, warning that changes to weights and configurations can weaken safeguards while task performance stays intact. A joint paper identifies 30 risk types across five categories and outlines a process for evaluating, mitigating and monitoring them, drawing on existing NIST and ISO standards. The paper is the first in a planned series on managing risks across open-weight AI deployments.

Mitigation

Verify model provenance with cryptographic signatures where available, and record unverified integrity as an open risk where they are not. Assess safety and capability separately, test the specific configuration to be deployed, reassess after material changes, and assess permissions, data access, connected tools, runtime behavior and operational controls at the application level. Gartner recommends independent adversarial testing, static scanning of model files, routing traffic through governed AI gateways, and isolating self-hosted runtimes.

Related items

• LowMicrosoft fixes classic Outlook bug that hid Copilot buttonsSame vendor · BleepingComputer
• InfoManaging open-weight model risk in enterprise deploymentsSame vendor · Microsoft Security Blog
• InfoCyberattack AIs must be trained only on air-gapped systems, hundreds of AI researchers saySame vendor · CSO Online
• MediumFake AI Attacks Spread Across US Companies: How MSSPs Can Stop Incidents EarlySame vendor · The Hacker News
• InfoAnthropic flags revenue surge to investors ahead of stock market debutSame vendor · BBC Technology