Recently led innovations in AI.
A selection of innovations on identity, access control, and zero-trust in agentic-AI I have recently led — from research and architecture to working systems, products, and patent-pending technology.
Zero-Trust in Agentic AI Systems
Agentic AI broke the rules of access control, the barrier enterprises must clear to reach production. I'm building a zero-trust layer that removes it: a new breed of AI-powered Auth Servers that stop agents from scope-breaking (exceeding what they're told to do).
Identity & Access Control in Agentic AI
You can't secure what you can't identify, and agentic AI just multiplied the unknowns. I designed the first open-source reference implementation of identity for agentic AI and co-led the Linux Foundation's AGNTCY Identity Working Group, giving agents and tools verifiable identities across the major IdPs (Okta, Entra, Ping, Duo, Keycloak, and Ory). It enables:
- Bring your own IdP; onboard devices with MFV
- Task/Tool-Based Access Control (TBAC), with or without human approval
Gluing AI, Data, and IAM On Premise
Compliance, data sovereignty, and cost keep enterprise data at the edge, so AI must move to the data, not the data to AI. I led the design of a new inference stack for Cisco AI servers that securely glues identity, language models, data, apps, and agents together at the edge:
- Identity-based "super-powers" (different users get different access to local data through the same chat interface).
- Onboard and link identity to local data sources in a few clicks (MCP, data connectors, tools, OpenAPIs)
- Bring Your Own Functions (BYOFs)
- Multi-modal use cases across text and images
Prompt Processing Units (PPUs)
I created PPUs — efficient units that characterize a prompt at inference time, detecting and distilling a set of features:
- The class of tasks requested to an AI model and/or agent
- The specific tasks within the class identified
- The data and parameters entered to complete the tasks
- Constraints that apply
- The desired output on completion
Transparent by design, PPUs deliver this feature as structured metadata in real time — with applications across security, Data Loss Prevention (DLP), observability, compliance, and productivity assessment.
→ Patent-pending technology