In an InfoQ talk, Manju Rajashekhar explains how the actor model can address the runtime problems of AI agents in production. Treating agents as distributed systems, the talk covers state ownership, message passing and supervision boundaries to make agent workflows resilient.
Symantec researchers have found an offensive AI agent that takes over unprotected language models online and uses them to attack servers, steal databases and install cryptominers. It can run up to 100,000 steps without human input, and the owners of the hijacked models are left paying the bill.
Dell is adding to its Dell AI Factory with NVIDIA lineup: three PowerEdge servers, a modular data center and workstations for building AI agents.
Vercel Labs has released fx, an experimental open-source coding agent written in Zig that can run as a CLI, a script component or an embedded library, including a WebAssembly build. Early feedback questioned its claim to be model-agnostic and flagged a name clash with an existing tool; Vercel says the pre-1.0 project is changing fast.
Tech companies pitch AI shopping agents as personal assistants once reserved for the wealthy. The Verge finds that, in practice, using one to buy things feels more like sending a nervous teenager to the store, while some retailers remain cautious.
On October 19, nine Chinese agencies issued an 18-point notice on commerce, one of which applies tiered governance to commercial AI agents: registration, testing and recalls in sensitive areas, self-assessment in low-risk ones. The guidance carries no penalties or timetable and does not say which tier these agents fall under; three days earlier, Beijing had proposed export controls on intelligent agents.