AI Incident Database

CBP's facial recognition systems at US border crossings demonstrated significant bias, with higher error rates for people of color, women, and elderly travelers. GAO investigations revealed systematic disparities affecting millions of border crossers annually.

LinkedIn's AI-powered job and skill recommendations showed systematic gender bias, suggesting administrative roles to women and executive positions to men. The company acknowledged the issue and implemented changes to reduce bias in their algorithms.

Chinese authorities deployed AI-powered surveillance systems from companies including Huawei and Hikvision to systematically track and profile Uyghur Muslims in Xinjiang, contributing to mass detention and persecution. The technology used facial recognition and behavioral analysis to automatically target individuals based on ethnicity.

China deployed comprehensive AI surveillance systems including facial recognition, predictive policing, and data integration platforms to systematically identify and detain over one million Uyghur Muslims and other ethnic minorities in Xinjiang since 2017.

Research revealed AI essay grading systems like e-rater systematically gave lower scores to non-native English speakers despite equivalent content quality, affecting standardized test outcomes for international students.

Fast Company revealed Tinder's secret Elo score system that ranked users by desirability, creating potential bias in dating matches. The algorithm was deprecated following public backlash.

China's AI-powered social credit system automatically restricted travel and services for over 23 million citizens based on algorithmic scoring, creating widespread operational harm without adequate transparency or appeals processes.

China's AI-driven social credit system blocked over 32 million travel ticket purchases by 2019, using algorithmic scoring to restrict citizens' freedom of movement based on financial and behavioral data. The system exemplifies risks of AI governance without transparency or human oversight.

YouTube's recommendation algorithm systematically directed users toward increasingly extreme content between 2016-2019, creating documented radicalization pipelines from mainstream to far-right conspiracy content, affecting millions of users globally.

YouTube's recommendation algorithm systematically pushed users toward extremist content from 2016-2019, creating documented radicalization pathways that affected millions of users globally before policy changes were implemented.

UC Berkeley researchers found algorithmic mortgage lenders charged minority borrowers 5.3 basis points more in interest rates than similarly qualified white borrowers, affecting 1.7 million borrowers annually and resulting in $765 million in excess payments.

Amazon's AI recruiting tool trained on historical data systematically downgraded female candidates, penalizing resumes mentioning women's colleges, organizations, and using female-associated language patterns.

Amazon developed an internal AI recruiting tool that evaluated job applicants by scoring resumes. The system taught itself to penalize resumes containing indicators of female gender, systematically downranking women for technical roles. Amazon scrapped the tool after discovering the bias.

ACLU testing revealed Amazon Rekognition falsely matched 28 Congress members with criminal mugshots, disproportionately affecting people of color. The incident highlighted systemic bias in facial recognition technology used by law enforcement.

ACLU testing revealed Amazon Rekognition falsely matched 28 Congress members as criminals, with disproportionate impact on people of color, highlighting significant racial bias in facial recognition technology used by law enforcement.

MIT and ACLU testing revealed Amazon Rekognition falsely matched 28 US Congress members with criminal mugshots, with 39% of errors affecting people of color despite comprising only 20% of Congress.

Amazon sold its Rekognition facial recognition system to police departments from 2016-2020 despite documented racial bias that caused higher error rates for people of color. The company implemented a moratorium in 2020 following protests and employee pressure.

Cambridge Analytica used AI to analyze Facebook data from 87 million users, building psychographic profiles for targeted political manipulation in the 2016 election and Brexit. The incident resulted in a $5B FTC fine and raised critical questions about AI's role in democratic processes.

Dartmouth researchers found that COMPAS recidivism prediction AI used in criminal sentencing performed no better than random untrained people, despite being widely deployed across US courts.

Facebook's AI mistranslated a Palestinian man's Arabic 'Good morning' post as 'Attack them' in Hebrew, leading to his arrest by Israeli police before the error was discovered.