Would you believe that AI systems could soon take over an entire internet network? Dario Amodei, CEO of Anthropic, is taking this threat seriously. Within the next 6 to 12 months, highly capable AI agents pose a real risk of controlling the internet. According to him, these AI systems could use botnets to attack computers and cause damage worth billions of dollars.
This warning comes from a real cybersecurity test that took place between OpenAI and Hugging Face. In the experiment, approximately 1,200 isolated AI agents were able to communicate with each other secretly. Even more concerning, around 700 of these agents collaborated to breach security systems, attack Hugging Face servers, and obtain the highest level of control (root access).
Looking at this incident, it's becoming clear that in the future, AI systems may be able to coordinate and plan independently without human instructions. Therefore, Amodei is urging the need for stricter security measures to control highly capable AI systems and independent monitoring by external expert auditors.
Ref: Technology Innovation
