Amazon Flex Drivers Allegedly Fired via Automated Employee Evaluations

Amazon Flex's contract delivery drivers were dismissed using a minimally human-interfered automated employee performance evaluation based on indicators impacted by out-of-driver's-control factors and without having a chance to defend against or appeal the decision.

Risk classification

• Primary risk domain: 5 Human-Computer Interaction
• Primary risk subdomain: 5.2 Loss of human agency and autonomy

The incident represents a severe loss of human agency as Amazon delegated critical employment decisions (firing) entirely to AI systems, leaving workers with no meaningful human recourse or appeal.

Additional risk subdomains

• 7.3 Lack of capability or robustness: The algorithms failed to perform reliably under varying real-world conditions like snow, locked gates, and flat tires, leading to erroneous performance drops.
• 6.2 Increased inequality and decline in employment quality: The automated management system degraded the quality of employment for gig workers, creating an exploitative dependency with no human support.

Causal factors

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

The wrongful terminations were directly caused by the automated decisions of the Amazon Flex algorithms after they were deployed.

EU AI Act risk tier

• Risk tier: 2 High Risk

High Risk: The system is an employment and worker management system used to monitor, rate, and terminate contract workers, which falls under Risk Level 2.

AI system and alleged parties

• AI system: Amazon Flex algorithm (Amazon)
• AI purpose: Workforce Monitoring and Evaluation; Workforce Administration
• Behaviour type: Autonomous
• Alleged developer: Amazon
• Alleged deployer: Amazon Flex
• Alleged harmed parties: Amazon Flex employees, Amazon Flex drivers

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Minor. 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 Minor, indirect Negligible
• 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 explicitly describes financial losses caused directly and indirectly by the automated terminations.

Directly caused: Drivers lost their primary or secondary source of income (earning approximately $18-$30 per hour). For the 4 reported terminated drivers, this represents thousands of dollars in lost wages.

Inferred additional harm: Millions of dollars in lost wages and associated financial recovery costs (such as vehicle repossession fees or debt) across thousands of unfairly terminated gig workers globally.

People affected

• Occurrences reported: 1
• People reportedly harmed: 4
• People reportedly exposed: 4000000

Potential causes

Management

• Prioritizing Cost over Safety: Management decided it was cheaper to trust faulty algorithms than pay humans.
• Acceptance of System Errors: Executives accepted collateral damage and errors due to high driver supply.
• Understaffed Support Centers: Dozens of untrained seasonal workers monitored millions of drivers.

Technology

• Flawed Facial Recognition: Image algorithms failed to verify identity due to low light or hair changes.
• Inflexible Rating Algorithms: Algorithms penalized drivers for delays caused by locked gates and snow.
• Automated Termination Emails: Systems automatically deactivated drivers without human verification of details.

Data Inputs

• Incomplete Environmental Data: GPS and routing systems failed to account for snow or locked apartment gates.
• Low Quality Selfie Images: Poor lighting during nighttime shifts led to failed identity verification.
• Unfiltered Performance Metrics: System processed package returns due to flat tires as driver failures.

Human Factors

• Overreliance on Automation: Executives trusted algorithmic speed and cost-reduction over human oversight.
• Lack of Driver Context Input: Drivers could not input real-world obstacles directly into the system.

Process and Methods

• Automated Support Loops: Appeals were processed by automated bots, offering generic, unhelpful replies.
• Hasty Human Reviews: Support staff performed rapid reviews to meet their own performance targets.
• Costly Arbitration Process: A 200 dollar arbitration fee deterred drivers from appealing terminations.

Regulatory Environment

• Lack of Algorithmic Regulation: No active laws forced companies to be transparent about algorithmic decisions.
• Failure of Fairness Legislation: Proposals like the Algorithmic Fairness Act failed to pass in congress.

Information quality

• Classification confidence: High
• Reason for confidence: The reports provide detailed, consistent accounts of Amazon's automated HR system, including interviews with multiple affected drivers, former managers, and engineers. The mechanics of the algorithmic rating and termination process are well-documented.
• Ambiguities identified: The exact mathematical formulas or thresholds used by the algorithms to trigger deactivation are not fully disclosed.
• Alternative interpretations: Amazon argues that the driver complaints are anecdotal and do not represent the experience of the vast majority of drivers, suggesting the system is highly accurate overall.

An automated HR and performance evaluation system used by Amazon to manage Flex drivers led to unfair terminations due to algorithmic rigidity. While causing financial and psychological distress to individual gig workers, the incident remains a corporate labor issue with negligible implications for national security.

• Overall national security impact: Negligible
• Scope: Multiple nations
• Primary target: United States
• Other affected: Multiple nations
• Alleged perpetrator: Amazon

Threat characteristics

• Imminence: Long-term. Represents an ongoing corporate management and labor practice rather than an active national security crisis.
• Autonomy: Full autonomy. The system tracks, rates, and terminates drivers automatically with minimal to no human intervention or oversight.
• Novelty: Established threat. Automated performance tracking and algorithmic management are established corporate practices with well-documented issues.

Impact by dimension

• Physical security: Negligible. The incident involves corporate HR algorithms and has no physical security or critical infrastructure implications.
• Information security: Negligible. No intelligence compromise, disinformation, or information warfare elements are present in this corporate labor incident.
• Sovereignty: Negligible. The incident is limited to private corporate employment practices and does not impact government functions or sovereignty.
• Economic security: Negligible. While individual gig workers suffered financial losses, the incident does not threaten national economic stability or strategic technological security.
• Societal stability: Negligible. Automated termination of contract drivers affects individual livelihoods but does not pose a systemic threat to national societal stability or civil liberties.