Institute for Decentralized AI launch

Cosmos Institute announced the Institute for Decentralized AI (IDAI) on September 15, 2025, with Samuele Marro as the Cosmos Incubation Fellow leading the launch. Marro is a DPhil student at Oxford's AIMS Centre for Doctoral Training, working on agent communication and collaboration with funding from Microsoft. He chairs the Lightweight Agent Standards Working Group.

The institute's research agenda has four named areas: trustworthy AI agents covering identity, authentication, monitoring, and accountability; anomaly detection in multi-agent language model networks with theoretical and statistical foundations; formalizing safety in agent networks via threat models and statistical relaxations; and trust and oversight protocols including reputation and mutual auditing. The technical position is that heterogeneous, distributed agent networks are the path to a more decentralized AI surface, and that the missing piece is the cross-agent protocol and oversight layer rather than the models themselves.

Marro's prior research includes Agora, a meta-protocol for inter-agent communication described in the arXiv paper "A Scalable Communication Protocol for Networks of Large Language Models" (2410.11905). Agora frames cross-agent communication as a trilemma between versatility, efficiency, and portability, and uses a mixture of structured routines, natural language fallback, and LLM-written routines to navigate it. This work sits in the same protocol-layer category as Anthropic's MCP and Google's A2A but explicitly targets heterogeneous, decentralized networks rather than a single hub-and-spoke runtime.

The institute's initial cohort consists of five fully funded research positions distributed between Oxford and Stanford. It sits at the protocols, sovereignty-decentralization, and governance cross-layers: it is upstream of any individual agent product, and the lock-in vector it is trying to push back on is the platform-agent monopoly Lazar describes in the same Cosmos cohort.