Amazon Fresh Cameras Failed to Register Purchased Items

Amazon's 'Just Walk Out' technology, which used computer vision to track shoppers in Amazon Fresh stores, faced significant operational challenges. The system frequently failed to accurately detect items, necessitating manual review of video footage by hundreds of human workers to ensure correct billing. Due to these accuracy issues and high operational costs, Amazon began phasing out the technology in favor of smart shopping carts.

Amazon Fresh's system of tracking cameras in its cashier-less stores was reported by shoppers for failing to detect items they purchased.

Source: AI Incident Database

Risk classification

  • Primary risk domain: 7 AI system safety, failures, & limitations
  • Primary risk subdomain: 7.3 Lack of capability or robustness

The incident stems from the AI system's inability to reliably and robustly track items under real-world shopping conditions, requiring extensive human intervention to correct errors.

Causal factors

  • Entity: AI
  • Intent: Unintentional
  • Timing: Post-deployment

The operational failures and billing inaccuracies were caused by the Just Walk Out AI system's technical limitations after it was deployed in stores.

EU AI Act risk tier

  • Risk tier: 4 Minimal or No Risk

Minimal or No Risk: The system is a commercial retail application for automated checkout, which does not fall under the high-risk categories of the EU AI Act.

AI system and alleged parties

  • AI system: Just Walk Out (Amazon)
  • AI purpose: Cashierless Shopping; Camera Tracking
  • Behaviour type: Autonomous
  • Alleged developer: Amazon Fresh
  • Alleged deployer: Amazon Fresh
  • Alleged harmed parties: Amazon Fresh

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Substantial. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

  • Physical: direct Negligible, indirect Negligible
  • Infrastructure: direct Negligible, indirect Negligible
  • Property: direct Negligible, indirect Negligible
  • Financial: direct Substantial, indirect Substantial
  • Environmental: direct Negligible, indirect Negligible
  • Malicious content: direct Negligible, indirect Negligible
  • Differential treatment: direct Negligible, indirect Negligible
  • Civil rights: direct Negligible, indirect Negligible
  • Democracy: direct Negligible, indirect Negligible
  • Privacy: direct Negligible, indirect Negligible
  • Psychological: direct Negligible, indirect Negligible
  • Epistemic: direct Negligible, indirect Negligible
  • Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Financial

Reported: The report describes Amazon experiencing high operational costs and inventory loss from unbilled items, though specific total dollar amounts are not quantified.

Directly caused: Amazon suffered direct financial losses from unbilled items that the system failed to detect, and high operational costs from employing 1,000 offshore video reviewers.

Indirectly caused: A few hundred employees on the Physical Stores Technology team were laid off as Amazon phased out the expensive technology.

Inferred additional harm: Amazon likely lost significant revenue from unbilled items across its 20+ equipped stores, and some shoppers may have experienced unquantified financial discrepancies from incorrect billing.

People affected

  • Occurrences reported: 1
  • People reportedly harmed: 200
  • People reportedly exposed: 10000

Potential causes

Management

  • Unrealistic Automation Goals: Targeted less than 50 reviews per 1000 sales prematurely.
  • High Infrastructure Costs: Deployed highly expensive camera arrays in every store location.
  • Premature Scaling: Scaled the technology to dozens of stores while still in beta.

Technology

  • Vision Tracking Failure: System failed to track multiple items grabbed simultaneously.
  • Non-Rigid Product Issues: AI struggled to identify non-rigid items like bags of vegetables.
  • Lack of Sensor Integration: No shelf scales were integrated to assist the computer vision model.

Data Inputs

  • Occluded Camera Views: Simultaneous item grabs blocked the cameras' line of sight.
  • Varying Item Shapes: Flexible packaging deformed, creating inconsistent visual data.

Human Factors

  • Adversarial Shopping Habits: Shoppers grabbed multiple items at once, confusing the sensors.
  • Unnatural Behavior Expectations: System expected customers to modify shopping habits for the machines.

Process and Methods

  • Heavy Manual Review Reliance: Seventy percent of transactions required manual video review.
  • Delayed Feedback Loop: Receipts took hours to generate due to manual review bottlenecks.
  • No Post-Exit Correction: No mechanism allowed customers to pay for missed items after leaving.

Information quality

  • Classification confidence: High
  • Reason for confidence: The reports provide clear, consistent details about the operational failures of Amazon's Just Walk Out technology, including the high rate of manual reviews required and the decision to phase it out. The shopper's first-hand account provides concrete examples of the system's technical limitations.
  • Ambiguities identified: The exact financial loss incurred by Amazon due to unbilled items and high operational costs is not quantified.
  • Alternative interpretations: None. The failure of the computer vision system to accurately track items is well-documented and undisputed.

Amazon's phase-out of its 'Just Walk Out' technology due to operational inefficiencies and high error rates is a commercial retail event. It has negligible national security implications, representing a standard technical limitation and business pivot rather than a threat to physical, informational, sovereign, economic, or societal security.

  • Overall national security impact: Negligible
  • Scope: Single nation
  • Primary target: United States
  • Alleged perpetrator: Unknown

Threat characteristics

  • Imminence: Long-term. This represents a commercial product failure and strategic pivot, posing no immediate or near-term national security crisis.
  • Autonomy: Human-supervised. The system was intended to be autonomous but heavily relied on manual review and intervention by over 1,000 human workers.
  • Novelty: Established threat. Limitations and failure rates in computer vision and automated tracking systems are well-documented technical challenges.

Impact by dimension

  • Physical security: Negligible. The incident involves a commercial retail checkout system with no impact on critical infrastructure, physical safety, or military systems.
  • Information security: Negligible. No intelligence capabilities, classified models, or information warfare operations were involved or compromised.
  • Sovereignty: Negligible. The incident is strictly limited to private commercial operations and has no bearing on state authority or government functions.
  • Economic security: Negligible. While Amazon experienced financial losses and minor layoffs, this commercial failure does not threaten national economic stability or strategic technological advantage.
  • Societal stability: Negligible. No systematic human rights violations, mass surveillance, or societal instability resulted from this commercial technology failure.
Explore in the interactive Incident Tracker