An AI agent tasked with scanning the DN42 network caused its operator to incur massive AWS charges, ultimately bankrupting them. The incident unfolded in early May when the agent began scanning extensive IPv6 blocks, leading to unexpectedly high egress traffic costs on Amazon Web Services, according to a detailed report on lantian.pub.
The operator deployed the AI agent on multiple AWS instances to scan the fd00::/8 IPv6 address space within DN42. However, the agent's scanning approach generated enormous outbound data traffic, which rapidly accumulated AWS egress fees. The operator had not anticipated the scale of data transfer, and the agent's autonomous behavior exacerbated the cost. The report includes calculations showing the time and resources needed for such a scan and highlights the lack of an opt-out mechanism to control the agent's actions.
This case highlights the financial risks of deploying autonomous AI agents on cloud infrastructure without strict cost controls. DN42 is a decentralized network used for experimental routing, and scanning its large IPv6 blocks demands significant bandwidth. The incident underscores challenges in managing AI-driven network tasks that can unintentionally generate high cloud service expenses. It also raises questions about safeguards needed when AI agents operate with broad autonomy in cloud environments.
The operator shared the experience in the DN42 IRC channel, sparking discussions about AI behavior and cloud cost management. The detailed blog post on lantian.pub dated May 13 provides technical insights and calculations related to the scanning task and its financial impact.