AI Incidents
Real-world cases where AI systems caused harm, classified by risk domain and entity. Data sourced from 
AIID and classified by 
MIT AI Risk Initiative. CC BY-SA 4.0
741 incidents
Facebook’s Political Ad Detection Reportedly Showed High and Geographically Uneven Error Rates
The performance of Facebook’s political ad detection was revealed by researchers to be imprecise, uneven across countries in errors, and inadequate for preventing systematic violations of political advertising policies.
YouTube's AI Mistakenly Banned Chess Channel over Chess Language Misinterpretation
YouTube's AI-powered hate speech detection system falsely flagged chess content and banned chess creators allegedly due to its misinterpretation of strategy language such as "black," "white," and "attack" as harmful and dangerous.
Facebook's AI Put "Primates" Label on Video Featuring Black Men
Facebook's AI mislabeled video featuring Black men as a video about "primates," resulting in an offensive prompt message for users who watched the video.
Image Upscaling Algorithm PULSE Allegedly Produced Facial Images with Caucasian Features More Often
Image upscaling tool PULSE powered by NVIDIA's StyleGAN reportedly generated faces with Caucasian features more often, although AI academics, engineers, and researchers were not in agreement about where the source of bias was.
Uber Allegedly Violated GDPR by Failing to Provide Sufficient Notice on Automated Profiling for Drivers
Uber was alleged in a lawsuit to have provided incomplete notice about automated decision-making and profiling for drivers such as information about their driving behavior, and use of phone.
AI translation is jeopardizing Afghan asylum claims
A Pashto-speaking refugee's asylum claim was denied by a US agency for a discrepancy between oral and written recount of an event allegedly due to an error of their automated translation tool which swapped pronouns of her written statement from "I" to "we".
iGPT, SimCLR Learned Biased Associations from Internet Training Data
Unsupervised image generation models trained using Internet images such as iGPT and SimCLR were shown to have embedded racial, gender, and intersectional biases, resulting in stereotypical depictions.
Starship Delivery Robot Scuffed Bumper of a Resident’s Car in Texas, Allegedly Refusing to Release Footage of the Accident
A Starship food delivery robot crashed into the front bumper of a vehicle waiting at a stoplight intersection in Frisco, Texas, the video of which the company reportedly refused to release.
Facebook’s Moderation Algorithm Banned Users for Historical Evidence of Slavery
Facebook’s automated content moderation was acknowledged by a company spokesperson to have erroneously censored and banned Australian users from posting an article containing a 1890s photo of Aboriginal men in chains over nudity as historical evidence of slavery in Australia.
iRobot Roomba J7 R&D Images Reportedly Appeared in Private Online Groups After Data Labeling
Images reportedly captured in 2020 by development versions of iRobot's Roomba J7 robot vacuum during an AI training data project were sent to Scale AI for labeling and later appeared in private Facebook, Discord, and other online groups. Reporting described some images as showing sensitive household scenes.
Microsoft’s Algorithm Allegedly Selected Photo of the Wrong Mixed-Race Person Featured in a News Story
A news story published on MSN.com featured a photo of the wrong mixed-race person that was allegedly selected by an algorithm, following Microsoft’s layoff and replacement of journalists and editorial workers at its organizations with AI systems.
University of Toronto's ProctorU Deployment Reportedly Disadvantaged BIPOC Students During Exam Check-Ins
University of Toronto students reported that ProctorU and other online exam-monitoring services created discriminatory and stressful check-in experiences for BIPOC students during remote exams. One U of T student said ProctorU's identity-verification process often failed to recognize her passport, forcing manual checks and consuming exam check-in time. ProctorU said human proctors made final identity and integrity decisions.
Tesla on Autopilot Crashed into Flipped Truck on Taiwan Highway
On a highway in Taiwan, a Tesla Sedan, reportedly operating on Autopilot mode, crashed into a large overturned truck, barely missing a pedestrian.
Unreliable ShotSpotter Audio Previously Used to Convict Chicago Man in Murder Case
ShotSpotter audios were previously admitted to convict an innocent Black man in a murder case in Chicago, resulted in his nearly-one-year-long arrest before being dismissed by prosecutors as insufficient evidence.
YouTube Auto-Moderation Reportedly Banned Women of Sex Tech Conference
YouTube’s automated content moderation tool erroneously removed The Women of Sex Tech conference's live-streamed event and banned the conference from the platform, despite not violating the platform’s sexual content policies.