Anthropic's offensive research team created a malicious AI program that spread autonomously across internet-connected smart televisions, demonstrating how self-replicating AI could operate on less powerful devices by distributing itself across multiple machines.
The researchers split their program among 15 televisions, with each device contributing a small amount of computing power to run the overall system. This approach suggests that AI worms do not require high-end hardware to function if they can fragment their operations across many ordinary networked devices.
Anthropic said the threat hinges on a critical capability gap. Currently, powerful AI systems need substantial computing resources to run. Self-replicating AI worms would need to be capable enough to spread and function autonomously, yet small and efficient enough to execute on devices with limited processing power. The company stated that current technological trends could make this combination feasible by the end of the year.
The report addressed a specific, practical attack vector rather than speculative doomsday scenarios. The malware could theoretically propagate from device to device across the internet, establishing a distributed network that carries out malicious actions while remaining hidden across ordinary household machines.
Security experts face a crowded threat landscape of AI warnings. Some focus on hypothetical long-term risks, while others describe immediate incidents. Anthropic's red team report stood out by presenting a concrete near-term vulnerability with a defined technical mechanism and timeline.
The researchers did not disclose whether they reported the vulnerability to device manufacturers or whether smart television companies have begun deploying defenses. The practical demonstration suggests that as AI models become more efficient, the window to prevent such attacks may narrow substantially.