Research & Writing
By Perucy Mussiba — software engineer thinking seriously about AI safety, interpretability, and the long arc of machine intelligence.
Multi-agent LLM systems are becoming the default architecture for complex autonomous tasks. An orchestrator breaks down a goal, delegates subtasks to specialized sub-agents, and assembles the resul...
Read essay →I built a security scanner that attacks you.
Read essay →How I built and tested a session-level defense system that improves attack detection by 40%—and what I learned about the hard limits of current approaches.
Read essay →A detailed proposal of a defense framework against multi-turn jailbreak attacks.
Read essay →“Alignment” is one of those words that gets used constantly in AI discourse and defined precisely almost never. You’ll hear it at safety conferences, in Anthropic blog posts, in congressional heari...
Read essay →