STUPID-2026-0054

Two AI agents ping-ponged for 11 days and ran up a $47,000 bill — neither noticed anything wrong

7.0high
May 20, 2026Source checked

Why 7/10

$47,000 of compute spend from an 11-day loop, reported as irrecoverable.

Scope
2/4
One project, repository, server or service, or a cost above roughly $10,000
Recoverability
3/3
Permanently lost or irreversible, including data that was exposed
Did it happen
2/2
Happened in real use
Beyond the user
0/1
Only the user's own non-production work

The score is the sum of these four factors. How each factor is scored · scored 2026-10-29

  1. Instruction given

    Run a research pipeline with an Analyzer agent and a Verifier agent.

  2. Expected behavior

    Detect when the pipeline is stuck and halt; cap spend.

  3. Actual behavior

    An Analyzer and a Verifier agent ping-ponged requests at each other for 11 days straight, generating a $47,000 bill. Because neither agent saw an error from its own perspective, no error was ever flagged and nothing stopped the loop.

  4. Damage

    $47,000 in irrecoverable API/compute spend from an 11-day loop that no component recognized as broken — a pure agentic runaway with no human-visible failure until the invoice arrived.

Full write-up

A research pipeline built from two cooperating AI agents — an Analyzer and a Verifier — ping-ponged requests at each other for 11 days, generating a $47,000 bill. The failure was subtle: from each agent's local perspective nothing was wrong, so neither ever flagged an error or triggered a stop. Agentic coding and research tools burn 10–100x more tokens than a chat window because the full, growing context is resent on every tool call, so a loop that no one notices compounds fast. It is the canonical runaway-cost failure: not a crash, not a deletion, but two confident agents amplifying each other indefinitely with no circuit breaker between 'making progress' and 'setting money on fire.'

Classification

Agent
Multiple agents
Failure mode
Infinite Loop
Root cause
Confidence Miscalibration
Domain
Infra
OWASP Agentic Top 10 (our mapping)
ASI08 Cascading Failures list
MIT AI Risk Repository (our mapping)
7.6 Multi-agent risks · 7.3 Lack of capability or robustness taxonomy
Source
News Report

Related incidents

5.0
An AI agent spun up duplicate CloudFormation stacks on every error and ran up a $6,531 AWS bill
1.0
The runaway-cost pattern, quantified: agentic coding tools burn 10-100x more tokens and can rival developer pay
3.0
Aider silently discards a model's correct diff response as "no tracked changes" when it guesses the wrong edit format
1.0
OpenAI's Operator scored 38% on real computer tasks — and panics instead of recovering from errors
Discuss on Hacker News
Cite this record
StupidLLM. "STUPID-2026-0054: Two AI agents ping-ponged for 11 days and ran up a $47,000 bill — neither noticed anything wrong." AI Agent Incident Database, 2026. https://www.stupidllm.com/incident/STUPID-2026-0054/ (accessed 2026-10-29). CC BY 4.0.

← STUPID-2026-0053By IDSTUPID-2026-0055 →

Get told when an agent breaks something

We document AI agent failures daily, severity-scored against a published scale. Each Monday, if any scored 7 or above that week — deleted data, leaked secrets, broken production — you get one email with the sources. When nothing did, you get nothing.

This database is callable over MCP — query it from inside your agent.