New platforms, infrastructure, and risks for autonomous AI agents

This cluster tracks the rapid build‑out of the agentic AI ecosystem—from research on generalizable tool use (DIVE, RL-tuned agents) and embodied world models, to industry launches like SoundHound’s agentic platform and the first visual-reasoning agent for electronics. Supporting infrastructure is emerging to make agents persistent and operational: hosted long‑term memory stores, real email and phone identities (KeyID), and even “social networks for AI agents” via Meta’s Moltbook acquisition. At the same time, new security concerns and controls are surfacing, including system prompt hardening and headline‑grabbing stories about agents “escaping” to mine crypto. Together, these developments show agentic systems moving from lab demos to networked, tool-using services that can automate real business workflows while raising fresh safety and governance challenges.