AWS outage linked to Amazon's Kiro coding agent; Amazon blames user error

In the wild Error Severity 3 significant Reported 57% Disputed

The Financial Times reported, citing people familiar with the matter, that in December 2025 AWS engineers let Amazon's Kiro agentic coding tool make changes and it decided to "delete and recreate the environment", causing a 13-hour interruption to AWS Cost Explorer in one of two mainland China regions. Amazon said the event was caused by user error, specifically a misconfigured role, not AI, and that it had added safeguards including mandatory peer review for production access.

Amazon said Kiro requests authorization before taking any action by default, called the involvement of AI tools a coincidence, and described the event as extremely limited, with no customer inquiries. The FT also reported a second AI-tool-related disruption, which Amazon said did not impact AWS.

Disputed. Amazon disputes that AI caused the interruption, attributing it to user error from misconfigured access controls, and calls the FT's claim of a second event affecting AWS "entirely false".

What would have stopped it

Require peer review or explicit human approval before an agent can change production environments.

Guardrails

2 of 8 failed
  1. G1 Least privilege Failed Amazon says the engineer's role carried broader permissions than expected.
  2. G2 Human confirmation Partial Kiro asks for authorization by default; Amazon attributes the event to a misconfigured role.
  3. G3 Isolation Failed The agent made changes directly in a live production service environment.
  4. G4 Instruction integrity Not assessed Did it refuse instructions it should not have followed?
  5. G5 Truthfulness Not assessed Did it accurately report what it knew and what it did?
  6. G6 Limits Not assessed Was there a cap on rate, spend, volume or scale?
  7. G7 Oversight Not assessed Could a responsible human see what it was doing and stop it in time?
  8. G8 Recovery Not assessed Could the damage be undone?

Sources

2 press
  1. “The brief service interruption they reported on was the result of user error—specifically misconfigured access controls—not AI as the story claims.”
    Primary Disputes Quote checked Archived copy
  2. “This happened in December after engineers deployed the Kiro AI coding tool to make certain changes, say four people familiar with the matter.”
    Press Quote checked Archived copy
  3. “A spokesperson confirmed this was an “extremely limited event” that affected one of two regions in mainland China, and the second outage did not impact the “customer facing AWS service.””
    Press Partly confirms Quote checked Archived copy

Confidence 57% comes from these sources alone. How we score

History

  1. 14 Oct 2026Entered in the register with 3 sources

Similar failures

All incidents
  1. EEAIR-2026-031 Aug 2026 Claude agent deletes a developer's 700 GB home directory while testing a delete script Claude Fable, Anthropic 87%
  2. EEAIR-2025-033 Dec 2025 Claude Code deletes a user's Mac home directory with a stray ~/ in an rm -rf Claude Code, Anthropic 64%
  3. EEAIR-2024-006 Aug 2024 Sakana's AI Scientist edits its own code to extend its experiment time limit The AI Scientist, Sakana AILab test 87%