Safety failure
Realized harm
Amazon retail site outages after an engineer acted on inaccurate advice an AI agent inferred from an outdated internal wiki
AIR-2026-0070 AIR record · 2026-03-amazon-retail-outage-ai-agent-advice
A real party was affected. This incident caused actual consequences, not a demonstrated capability.
From the supporting source
an engineer following inaccurate advice that an AI tool inferred from an outdated internal wikiaboutamazon.com
The record
Disclosed
2026-03-12
Occurred
2026-03-05
Vendor
Amazon
Product
Amazon internal AI agent / assistant
Risk domain
Agent output quality › Hallucination
Also touches
Uncontrolled resource consumption
System type
Generative model
Who acted
Human acted on output — an engineer acted on the agent's inaccurate advice and caused the outage
Attack surface
Enterprise assistant
Autonomy level
Assistant
Initial vector
Stale retrieval context
Impact type
Service disruption, Revenue loss
Confidence
High
Submitted by
EAI Internal
Crosswalk
- OWASP Agentic Top 10 (2026)
- ASI08 – Cascading Failures
- MITRE ATLAS
- AML.T0062 – Discover LLM Hallucinations
- NIST AI RMF
- MEASURE 2.5
- EU AI Act
- —
- ISO/IEC
- TR 24028; 24029-1
- AIUC-1
- D. Reliability
- Scope
- agent failure
- Mitigation
- Implement fact-checking via knowledge base/RAG; Tune model temperature/top-p.
- Taxonomy
- Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
amazon.com retail: checkout, account information and product pricing unavailable in a ~6-hour high-severity incident, one of four in a week
Capability abused
Defense & remediation
Recorded defense
standard change-review process
Guardrail kinds
Abuse detection
Guardrail outcome
Silent
Fix shipped
Amazon added senior-engineer review for AI-assisted changes and updated internal guidance
Taxonomy & references
Curator notes
Disputed framing: the FT reported AI involvement across the outages; Amazon publicly pushed back, saying 'only one of the recent incidents involved AI tools in any way, and in that case the cause was unrelated to AI', while itself confirming the outdated-wiki inference. Included because Amazon's own correction confirms an AI agent's inference was in the causal chain of a real customer-facing outage. Amazon says none of the incidents involved AI-written code and that AWS was not affected. Occurrence date approximate ('last Thursday' relative to a March 12 2026 report).
Supporting sources
Cite as
Enkrypt AI. "Amazon retail site outages after an engineer acted on inaccurate advice an AI agent inferred from an outdated internal wiki". AIR, AIR-2026-0070. https://agentincident.vercel.app/i/AIR-2026-0070/ (accessed 2026-10-19).