Some work I’ve done ヽ(o^ ^o)ﾉ

Current focus: AI red teaming

My research now centers on the security of LLM applications and agents. I examine how retrieved content, messages, and tool responses can cross trust boundaries, influence decisions, or reach privileged actions.

• Prompt injection: direct and indirect instructions that redirect an application’s intended behavior.
• RAG and data exposure: failures in retrieval boundaries and access to sensitive information.
• MCP and tool security: tool permissions, authorization checks, and trust between agents and connected services.
• Vulnerability validation: reproducible proofs of concept, attack-chain analysis, and root-cause reporting.

Current research profile

Earlier work: blockchain security

These projects and reports document my previous focus on smart contracts and protocol security. Public audit results include Rain and stNXM (each 1 Critical and 1 High), and BattleChain Confidence Pools (1 Medium). The archive records bug-bounty outcomes separately, including duplicates.

Read the audit reports · View my Sherlock profile