Independent security researcher. I work on containment and verification of autonomous AI agents, cryptographic interlocks, and reproducible security research.
Research is published openly under CC BY 4.0 with code under MIT, and every paper has a public repository of reproducible artifacts. This work was conducted independently and is not affiliated with or representative of any employer.
| Paper | DOI | Artifacts |
|---|---|---|
| NIGHTCRAWLER: Temporal Causal Closure for Persistent and Deferred Effects of Autonomous AI Agents | 10.5281/zenodo.23265287 | nightcrawler |
| The Agent Security Control Plane: Toward a Zero-Trust Architecture for Autonomous Machine Cognition | 10.5281/zenodo.23146801 | agent-security-control-plane |
| Containing Cyber-Capable AI Agents | 10.5281/zenodo.23124432 | ccaa |
| Memory-Egress Cryptographic Interlock (MECI) | 10.5281/zenodo.23109676 | meci |
| Attack Calculus: A Typed, Evidence-Aware State-Transition Calculus for Cross-Domain Cybersecurity Reasoning | 10.5281/zenodo.23092790 | attack-calculus |
| Mission-Invariant Architecture Morphing (MIAM) | 10.5281/zenodo.23001045 | miam |
- llm-agent-control-plane: defensive reference implementation that keeps authorization, policy, approval, and audit logging outside the model for tool-connected LLM agents.
- model-identity-verifier: command-line tool that checks whether a language model identifies itself consistently and flags route, provider, or downgrade signals.
- rte-a-reference: reference implementation of the Red Team Engineering Algorithm with attributed tasking, tamper-evident audit logs, and ephemeral infrastructure.
- db-access-lab: disposable PostgreSQL lab for evaluating third-party products that request database access.
- dockercomms: OCI-native secure file transport with signed artifacts.
- Research lab and archive: codethor0.github.io/thor-sec
- ORCID: 0009-0001-6573-385X
- Security reports: please use the private vulnerability reporting option on the relevant repository.