One AI agent made 15,000 high-cost calls in an hour, racking up $50,000 in cloud charges
shashib · x · 2026-10-10
Citing a Mandiant story, the author describes a company's own AI agent that made over 15,000 high-cost calls in under an hour and ran up roughly $50,000 in cloud charges — with no attacker involved.
Key points:
- Most security spending points outward, but the inside is more dangerous: we relax around anything we've already approved
- An agent sits inside the wall with credentials and a budget; nobody reviews a colleague's every call, so nobody reviewed this one's
- "An agent without a strategy is an insider with your credit card and no supervisor"
Before production, four questions: Who owns it by name? What can it call? What's its spend ceiling? Who can shut it off, and how fast?
Related event: Runaway AI Agent Burns $50K in Cloud Costs in One Hour(2 posts)→
More from coding & agent
- Salesforce's Opera critic framework boosts coding agent resolve rates by up to 15 points — Salesforce · 2026-10-10
- Prime Sandboxes 向 A 轮前初创免费开放 RL 训练沙箱额度 — xeophon · 2026-10-10
- Why OpenAI built DIL instead of rendering generated HTML in iframes — soleio · 2026-10-10
- Jev inside agents: browser-use's jev-ultrafast hits 22.5k stars picking actions without LLM calls — Arindam_1729 · 2026-10-10
- Cloudflare gives you 1,000 free outbound emails a day — just have your agent install the CLI — tristanbob · 2026-10-10
- Routing 1000+ video models: filter with code first, let the LLM pick from a shortlist — Vaciuum1 · 2026-10-10