AI Research Roundup: Agent Behavioral Contracts, Autonomous Memory, and Certified Circuits
Three new papers tackle agent reliability through formal contracts, active knowledge acquisition for memory systems, and provably stable mechanistic interpretability.