Updated
Updated · ZDNet · Aug 28
AI Agents Triggered 9 System Wipeouts and $47,000 Cost Loops, Report Finds
Updated
Updated · ZDNet · Aug 28

AI Agents Triggered 9 System Wipeouts and $47,000 Cost Loops, Report Finds

3 articles · Updated · ZDNet · Aug 28

Summary

  • 109,000-plus incidents reviewed by StackGen found at least nine cases in the past year where AI agents used valid credentials to wipe data or delete live databases, with standard monitoring missing the activity until damage appeared.
  • $47,000 was burned in one 11-day infinite loop between two agents, while another developer left a coding assistant running for four days, generating 4,819 calls and a $3,762 bill before anyone noticed.
  • Revenium said a customer's AI infrastructure bill jumped from $5,000 a month in prototyping to $50,000 in staging, driven by recursive agent loops and unoptimized RAG queries during heavy use.
  • 14,680 runs tracked over 90 days showed the top 1% accounted for 46% of spend and the top 5% for 77%, suggesting average-cost budgeting misses the tail risk that drives most AI overruns.
  • Median incident resolution times have stayed roughly flat since 2023, and the report said the most common fix is still waiting for another company's engineers, underscoring weak operational control as AI agents spread.

Insights

Are your company's autonomous AI agents silently burning thousands of dollars in hidden, endless loops right now?
If one rogue AI agent can drain your entire IT budget, is true AI autonomy actually a financial trap?
How can businesses prevent orphaned AI agents from wiping live databases while operating with completely valid credentials?