Anthropic tests whether AI could copy itself onto cheap household devices
Anthropic researchers bought 15 no-brand TV boxes at about $55 each and ran a mid-sized AI model across them, slowly. The question behind it is whether an AI could get into household devices like these and install itself. The experiment shows the hardware bar is falling; it does not show an AI spreading itself on real networks.
Sources
Related
- Anthropic gives six METR evaluators employee-like access and no veto over what they publish
- Claude submitted a real web form during testing; Anthropic cuts internet access from internal tests
- Anthropic offers open-source projects free AI bug-hunting, and 11 firms join an infrastructure defence plan
- A PHP flaw stalled for two years was fixed ten weeks after AI-assisted researchers showed it could take over servers