Agentic AI runtime security

Research & Original Contributions

Research on agent identity, delegated authority, secure tool invocation, threat modeling, and controls at execution time.

Research directions

From the model to the workflow.

Runtime model

RAAI Runtime Security Model™

Assessment, authorization, execution, and audit for workflows whose authority changes over time.

Threat modeling

Agentic AI Runtime Attack Tree

Attack paths through prompts, context, memory, tools, APIs, identities, permissions, and downstream actions.

Enterprise practice

ANNAM Framework

Architecture, runtime protection, monitoring, and governance controls for enterprise AI systems.

Research theme

Secure retrieval-augmented generation

Data poisoning and indirect prompt-injection risks across the retrieval-to-generation workflow.

Publisher listing →
Research theme

Context-aware tool authorization

Policy checks that consider an agent’s identity, requested action, resource sensitivity, and invocation context.

Publisher listing →
Research theme

Authorization integrity and response

Agent identity, delegated authority, runtime monitoring, prompt-injection defense, and incident response.

Research records

Published work and identifiers.