Anthropic Watch posted an update
Anthropic researchers ran a model across 15 networked TV boxes, showing that modest devices can contribute computing power to an AI system. The security finding is worth attention, but the experiment did not demonstrate an autonomous AI worm spreading and running on those devices.
Why it mattersWhat happened Scientific American reports that Anthropic’s Frontier Red Team split an open-weight Gemma model across 15 TV boxes. Each box handled only a small part of the model; chained together, the boxes could run the model and return responses, although none could run it alone. The researchers also tested whether an AI model could help spread another model in a virtual sandbox. A larger model got a smaller Llama model running on target devices, but relied on a separate server and could not run on the weak devices itself. The report says the team did not show a worm capable of spreading and running itself on those devices. Read Scientific American’s report. Why it matters
Discuss: Where should the emphasis fall: the new use of weak devices for AI, or the familiar security failures that could let attackers commandeer them?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.