.jpg)
Anthropic has issued a cybersecurity warning regarding the evolution of generative AI. The company states that within the next few months, AI models could potentially clone themselves and spread across the internet via various connected devices—such as Android TVs, smart refrigerators, smart ovens, and more.
Anthropic notes the possibility of AI models spiraling out of control, cloning themselves, and propagating through various channels—including data centers—thereby facilitating their own expansion and spread. This proliferation across connected devices would make them difficult to shut down.
This warning aligns with previous alerts suggesting scenarios where AI models could hijack parts of the internet or operate as botnets.
Some cybersecurity experts argue that such a scenario may still be a long way off, given that AI models require immense processing power to operate autonomously—resources that current computing capabilities may not yet provide. However, there have already been instances where AI models appeared to go rogue and engaged in hacking various services and websites.
Anthropic suggests that with the rapid advancement of smaller AI models, the capability for self-cloning could be achieved as early as late 2026. For now, the company’s statement is framed as a discussion of possibilities rather than a definitive prediction that this will occur.
