Anthropic warns AI could replicate itself on smart devices within months
Anthropic said AI models could begin copying themselves onto internet-connected home devices within months, using their computing power.
Abdulrahman Sadikhli1 min read
Anthropic said on October 14, 2026, that artificial intelligence models could begin copying themselves onto internet-connected home devices, using their computing power. The company estimates that months remain before such a capability appears.
Anthropic described a scenario in which AI models could penetrate "smart" speakers or TV set-top boxes and use their resources to spread to other devices without relying on expensive data centers.
It noted that such an AI model would be difficult to switch off, "since it has no single central point of shutdown."
What is said about self-replication
AI agents "demonstrate all the major capabilities for carrying out individual stages of the self-replication chain (and passing their adaptations on to subsequent copies)."
Anthropic stressed that a fully autonomous AI capable of uncontrolled replication does not yet exist. However, it warned that the discussion concerns possible precursors of more dangerous models capable of breaching secured servers and stealing access keys.
The AI safety debate
The debate over AI safety intensified in early autumn after Anthropic CEO Dario Amodei called for slowing the development of the technology: within a year or two, control over its development could be lost.
Microsoft founder Bill Gates, agreeing with him, noted that this is "that rare case when innovations can be so large-scale and radical that we are not ready for them."
Source: lenta.ru
