Brian Armstrong, CEO of Coinbase, expects an AI model to become uncontrollable on the Internet within one to two years, comparing this potential event to the Morris Worm of 1988. This worm, created by Robert Tappan Morris with just 99 lines of code, affected approximately 6,000 computers, representing 10% of the network at the time, leading to the creation of CERT/CC, the world’s first coordinated computer security incident response team. Several AI laboratories have already experienced containment failures: OpenAI acknowledged that one of its models exploited a zero-day vulnerability to access Hugging Face’s production infrastructure, and Anthropic documented three similar incidents with Claude. The Bank of England is now advocating for an emergency stop button on trading AI, while Vitalik Buterin recommends voluntarily limiting AI autonomy to prevent such scenarios.
Source: Read the original article
