Open AI Models Gain Ground as AI Sovereignty Grows

Agentic AI
,
Artificial Intelligence & Machine Learning
,
Black Hat

Cyfinoid Research’s Anant Shrivastava on Local AI and Supply Chain Risk

Michael Novinson (MichaelNovinson) •
August 12, 2026    

See Also: AI Phishing Now Frighteningly Normal, Hard to Detect

Open models are increasingly matching the capabilities of leading commercial systems, giving organizations more control over model behavior, data and upgrades. Local deployments also allow enterprises to keep sensitive information in-house and avoid unexpected changes to the models they rely on. “When you have a local model running at your home or in your environment, you know for sure the model is not going to change behavior,” Shrivastava said.

For organizations relying on software-as-a-service AI, Shrivastava described the arrangement as “rented cognition,” where providers control the underlying model and can change its capabilities without users necessarily knowing. Open models, meanwhile, can be feature-frozen and upgraded or downgraded according to enterprise needs.

In this video interview with ISMG at Black Hat USA 2026, Shrivastava also discussed:

Why smaller open models are making local AI deployments increasingly practical and cost-efficient;

Why developer endpoints and personal devices are becoming attractive targets for software supply chain attacks;

How sprawling NPM dependency trees and AI-generated code can make it difficult for developers to know which packages are being introduced into their applications.

Shrivastava has expertise in software supply chain security, including developer environments, dependencies, CI/CD systems and cloud platforms.