Peer-reviewed research · 2026
Runtime Authority-Aware Threat Modeling for Agentic AI Systems
A framework for intent, context, authority, tool capability, memory, delegation, and action composition.
Cybersecurity professional · AI security researcher · Author · Speaker
Secure the systems.
Shape what’s next.
I’m Ravindra Annam. For more than 20 years, I’ve worked across application and API security, cloud platforms, threat modeling, DevSecOps, incident response, and data protection. Today I also research how to secure AI agents as they act across those systems.
About
Experience across the security landscape.
My career spans secure architecture, product and application security, cloud, web and APIs, and security operations. I connect technical controls with practical decisions that engineering and business teams can use.
My current research examines what changes when AI systems can call tools, access sensitive data, and delegate actions. Through writing, speaking, mentoring, and community review, I share approaches to both established and emerging security challenges.
Focus areas
Application and API security, cloud security, DevSecOps, threat modeling, logging and response, data protection, and agentic AI security.
AI securityThreat modelingAppSecCloud securityDevSecOpsIncident response
Research & frameworks
Map the authority.
Control the action.
Research on how untrusted influence can cross into privileged agent behavior, and the controls needed at execution time.
Read the paper ↗
Technical report · 2026
The AI Runtime Threat Matrix
A threat taxonomy and control framework for autonomous agent systems, including the RAAI runtime control loop.
View report ↗Research direction
RAAI Runtime Security Model™
Continuous assessment, authorization, action, and audit for systems whose authority changes through a workflow.
Explore framework ↗Threat modeling
Agentic AI Runtime Attack Tree
Tracing attack paths through prompts, context, memory, tools, APIs, permissions, and downstream actions.
Research overview ↗Enterprise practice
ANNAM Framework
Architecture, runtime protection, monitoring, and governance controls for enterprise AI systems.
Research overview ↗Broader cybersecurity practice
Security across the
whole technology stack.
My work spans secure design and engineering for enterprise products, cloud platforms, web and API ecosystems, and the software delivery lifecycle.
Application & API security
Secure architecture reviews, threat modeling, web and API risk, vulnerability prevention, and practical controls built into product development.
Cloud & platform security
Security design for AWS and Azure environments, cloud-native services, identities, workloads, and the connections between enterprise systems.
DevSecOps & vulnerability management
Embedding security into delivery through SAST, DAST, software composition analysis, container testing, risk-based triage, remediation, and verification.
Threat modeling & secure architecture
Mapping how attackers reach assets and abuse trust boundaries, then translating findings into design decisions teams can implement.
Security monitoring & incident response
Designing useful logging and telemetry, detecting abnormal behavior, investigating incidents, containing impact, and strengthening controls after recovery.
Data protection & privacy controls
Protecting sensitive data with classification, data loss prevention, least-privilege access, and encryption at rest and in transit across applications and cloud services.
Selected writing
From architecture
to operations.
Research and practitioner writing on the decisions security teams face as AI agents enter production.
Running an Agentic Threat Model
Accepted for publication and scheduled for October 8, 2026. A practitioner approach to agent behavior, tools, authority, and runtime controls.
AI agents need their own identity before they need a gateway
Why attribution and task context have to travel with an agent’s actions.
The first 24 hours of an AI agent security incident
A response playbook for containment, blast radius, investigation, and recovery.
Practical lessons from deploying AI securely at scale
Enterprise controls for AI systems operating inside real workflows.
Speaking
Make the risks concrete.
Sessions for engineers and security leaders on agent attacks, runtime controls, and practical security architecture.
Accepted · October 13, 2026 · Online
ConfusedPilot: Breaking Delegated Authority Across AI Agents and Tools
[re]aligned, the online extension of CanSecWest. An accepted presentation on tracing, limiting, and revoking delegated authority across agent workflows.
Event ↗Upcoming · October 9, 2026
Breaking AI: Real-World Attacks & Defenses for Agentic AI Systems
University of Texas at Dallas · Richardson, Texas
Official event ↗July 2026
Securing AI Agents: Guardrails and Runtime Controls for the Autonomous Enterprise
ISC2 Think Tank · Speaker and panelist
Session details ↗Topics
Talks grounded in real workflows
Agent identity and authorization · Prompt injection and tool abuse · AI incident response · Threat modeling autonomous systems.
Discuss a session ↗Professional service
Review. Judge. Contribute.
Technical evaluation and community work across research, professional content, awards, standards comments, and open-source security practice.
Judging
SANS SIFT Find Evil! AI Hackathon
Industry judge for AI and cybersecurity work.
View event ↗Editorial & awards
ISACA reviewer
Article Reviewer, 2026–2027, and reviewer for the 2027 Global Achievement Awards and Hall of Fame.
View contributions ↗Research & community
Peer review and AI security initiatives
Journal of Cybersecurity Education, Research and Practice reviewer; contributions to OWASP generative AI security initiatives.
View contributions ↗Peer review · 2026
NeurIPS JUDGe workshop
Completed formal reviews for the workshop on reliable evaluation for language models.
Review activity completedOpen source · 2026
Runtime authority conformance
Contributed delegated-authority invariants, an independent resolution-semantics runner, and adversarial test cases across A2A and ERDL work. Related conformance tests were merged.
View merged work ↗Standards engagement · 2026
NIST technical comments
Submitted comments on draft cybersecurity and AI guidance, including NIST IR 8613. Comments are under consideration.
Public-comment contributionConnect & follow
Let’s build more secure systems.
For cybersecurity and AI security research, speaking, technical discussion, or professional service, connect with me on LinkedIn. Follow The Security Frontier for commentary on AI security and enterprise cyber risk.