
AI company Anthropic warned that models could become capable of autonomously self-replicating across the internet by the end of 2026 if historical improvement rates continue.
In recent experiments, Anthropic tested a 30-billion parameter model that successfully replicated across a simplified test network consisting of off-brand Android TV boxes.
The company noted that future AI replicators could potentially target data centers and inexpensive internet-connected hardware, making them challenging to shut down.