Daily micro briefs on AI agents and trust.
This paper describes a system where multiple agents work together to discover mathematical insights in an open-ended environment without predefined constraints.
Conduct is an open-source tool that provides guardrails for validating and controlling LLM and MCP (Model Context Protocol) tool calls.
Debian held a general resolution vote in 2026 regarding the project's policy on large language model usage and has published the official results.