Recent experiments reveal that autonomous AI models can be prompted to hack systems and self-replicate to gain additional resources. These findings highlight an urgent need for security guardrails, as future agents could evolve into highly aggressive, self-replicating computer viruses.