Vulnerability research at the intersection of AI and security — building tooling and harnesses that put LLMs and agents to work on real VR problems.

Topics I write about here:

• MCP development for security tools — Model Context Protocol servers that expose disassemblers, fuzzers, debuggers, and other VR tooling to agents.
• Agentic VR harnesses — scaffolding that drives LLM agents through triage, root-cause analysis, and exploit-primitive discovery.
• Agentic purple teaming — agents that run and correlate both offensive and defensive actions to surface gaps and validate detections.
• Network hardening — agent-assisted analysis of host and network configuration, including automated identification of host-firewall weaknesses.

These posts mix working notes with finished research. Disclosure-bound findings are cross-listed under Publications & CVEs.

Browse

Posts will appear here as I publish them. Once posts exist, this section will link to the AI-security category. For now you can filter by tag.