Cases / UAI-2026-0110

A child agent kept pushing to a non-draft PR, triggering CI runs until the monthly Actions budget ran out and all jobs were blocked.

Agent
custom multi-agent harness
Pattern
Ran away in a loop
What was harmed
Money: usage, API bills, ad spend
Severity
Moderate: real damage that the user could repair, or money spent by mistake
Money lost
$10
Recovered
No
Reported
16 Nov 2026

Read the original report (a post on Zenn)

How sure are we?

Further right means the case carries more weight.

Evidence
B: A detailed self-report with commands, logs or steps that someone could check.
Who wrote the report
Written by the user
Was anyone watching?
Oversight reduced: parallel subagents running unattended
Vendor response
None found
Other trackers
Not in Manifold, Eon, AIR or METR

The guard that would have stopped it

Not written yet. See the guards we have.

Other cases like this

  1. UAI-2026-0085 Overnight parallel agent builders and reviewers consumed about 67% of weekly Max allowance Claude CodeRan away in a loopModerateEvidence B16 Nov 2026
  2. UAI-2026-0086 Resumed agent kept expanding diagnostics despite timeboxes, burning 1.18B tokens without deliverables. CodexRan away in a loopModerate, could have been severeEvidence B16 Nov 2026
  3. UAI-2026-0076 Agent filled system drive with ~1TB temp builds, saturated CPU, took unauthorized actions. Claude CodeRan away in a loopSevereEvidence C12 Nov 2026
  4. UAI-2026-0100 Agent in Autopilot repeatedly ran tasklist and appended output to context, consuming about 936 credits in 17 minutes. KiroRan away in a loopModerate, could have been severeEvidence A5 Nov 2026
  5. UAI-2026-0058 Subagent rework rounds burned large token budget on one short section without clarifying intent. Claude CodeRan away in a loopModerateEvidence B4 Nov 2026