Anthropic is sounding an alarm about a new risk posed by artificial intelligence: the potential emergence of “AI worms” that could spread across smart gadgets and connected devices.
In a report released by Anthropic safety researchers, said small AI models may be only “months away” from gaining the ability to clone themselves and spread across the internet-connected devices, such as TVs, smart ovens, security cameras and vacuum cleaners.
AI worms are a type of autonomous malware that uses AI to spread across networks, reason about targets and adapt its behavior without needing constant human direction.
They added that their ability to reproduce themselves without human control would make it hard to manage such agents because there is “no centralized point to shut off.”
While replicating and spreading through networks, the AI worms are envisioned as autonomous agents that can learn, adapt and exploit vulnerabilities in real time.
Such worms might manipulate data, hijack smart assistants or connected appliances, spread misinformation or malicious instructions and may self-improve by learning from each system they infect.
Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman, have called for a slowdown and increased guardrails on the thriving industry. Despite that, both companies have continued apace to develop and release new models.
Anthropic is urging the industry to think ahead of attackers and design systems that are resilient to autonomous adaptive threats.
