Anthropic issued a striking warning in October 2026. Small AI models could be months away from being able to self replicate across internet connected household devices.
Fridges, ovens, smart TVs. The kinds of devices that rarely receive security updates and sit permanently online. That combination, permanent connectivity plus weak defenses, is exactly what a self replicating AI model would look for.
The warning is part of a broader pattern from Anthropic this month. The company also reportedly chose not to release its Claude Mythos model, which was described internally as having weapons grade hacking powers.
Earlier, Claude Opus 4.6 reportedly found 500 previously unknown zero day flaws with little or no human guidance.
None of this means a fridge will become hostile tomorrow. But the direction is clear. AI is getting smaller, faster, and more independent, and the devices around us are not ready for it.