Agent Control Standard
Runtime governance for AI agents
AI security,fromresearchto working systems.
I investigate emerging threats, build products and tools from 0→1, and help turn the findings into open standards and practical guidance.
Pillar Security · OWASP GenAI Security Project
Selected work ↓01ResearchFind and characterize the risk
02BuildTurn evidence into a working system
03StandardizeMake the learning useful to others
Selected work
Research. Build. Standardize.
01
Security research
Inference-Time Backdoors via Hidden Instructions in LLM Chat Templates
Co-author
02A new attack surface in the open-weight AI supply chain, evaluated across 18 models and 3,868 agent episodes.
Read the paper ↗Applied security
Pillar Agentic Security CTF
Creator
03A hands-on environment for learning how agentic systems fail—and how defenders can reason about those failures.
Enter the CTF ↗Open standard
Agent Control Standard
Specification co-lead · OWASP GenAI initiative
A vendor-neutral standard for intercepting, observing, and governing AI agent actions at runtime.
Explore the standard ↗Open work
OWASP GenAI
Current rolesI help move agentic security from emerging research toward shared language, practical guidance, and interoperable controls.
Specification co-lead
State of Agentic AI Security & Governance V2
Safety, security, taxonomy, and governance
Securing Agentic Applications Guide V2
Agentic hooks
Security findings
Vulnerabilities disclosed and fixed.
CVE2026-45792
GHSA-fvvm-949w-qj4w · Reporter
RTK trusted project-local filters that could silently alter command output shown to an LLM.
CVSS 6.9 · ModeratePatched in 0.32.0Read the advisory ↗
Writing
Writing across research and practice.
01
Browse all writing →From a Copilot to Cluster Admin: inside AtlasOps, our free agent-security CTF
Pillar Security
↗02The Fable Recall Puts the Spotlight in the Wrong Place
Pillar Security
↗03Your agents answer to Hades: how one commit hijacks 4 AI coding tools
Pillar Security
↗04Standardizing the Control Plane for AI Agents: Pillar's Role in ACS v0.1.0
Pillar Security
↗05Untrusted Project-Local Filters in RTK: When Your AI's Eyes Are Someone Else's to Control
Pillar Security
↗Publications
A research record across AI security and learning sciences.
Complete list ↗2026↗2023 2021 2016 2015
Inference-Time Backdoors via Hidden Instructions in LLM Chat Templates
ICLR 2026 Trustworthy AI Workshop · arXiv:2602.04653
Pose Detection: Towards Ubiquitous Embodied Interventions
International Conference of the Learning Sciences
Directed Epistemic Network Analysis
International Conference on Quantitative Ethnography · Best Student Paper
Pre-Reform Health Care Access and Affordability within the ACA's Medicaid Target Population
Urban Institute Health Policy Center Report
Primary Care Providers Ordered Fewer Preventive Services for Women with Medicaid than for Women with Private Coverage
Health Affairs · 34(6), 1001–1009
About
Bridging research and industry.
I’m a software engineer and researcher with a background spanning learning sciences, open-source tools, technical founding, and AI security.
At Pillar Security, I build products from 0→1 while researching the threats shaping the next generation of AI systems. I’m most useful when the question is unresolved and the answer needs to work outside the lab.
Research collaborations · Standards · Early product work