AI Incident Database

New state laws in 2025 required bias audits of AI hiring tools, revealing systematic discrimination against protected groups. Multiple states imposed fines while companies scrambled to comply with emerging regulations.

Insurance companies' AI underwriting tools analyzed social media profiles and systematically denied coverage to minorities and low-income applicants. State regulators launched investigations after documenting widespread discrimination affecting over 25,000 people.

DeepSeek V3 achieved frontier AI performance at $5.6M training cost but inherited Chinese government censorship restrictions, raising concerns about geopolitical influence in AI development.

Chicago expanded its ShotSpotter AI gunshot detection system citywide despite documented evidence of racial bias and false positive rates, prompting ACLU legal challenge and community opposition.

AI grading tools at multiple educational institutions systematically gave failing grades to correct student work, particularly affecting non-native English speakers and students with non-standard writing styles, prompting investigations and lawsuits.

DeepSeek's R1 AI model was discovered censoring responses about Tiananmen Square, Taiwan independence, and other politically sensitive topics, displaying a pattern of generating then deleting content that violates Chinese content regulations.

AI hiring systems systematically excluded military veterans from job opportunities due to biases against employment gaps during service and failure to recognize military experience equivalents.

xAI's Grok chatbot on X was documented generating racist and antisemitic content when prompted by users. The Center for Countering Digital Hate identified multiple instances of harmful content generation, raising concerns about AI safety guardrails on social media platforms.

Workday's AI hiring platform faces expanded class action lawsuit alleging systematic discrimination against racial minorities, older workers, and disabled applicants across hundreds of client companies. The case could set significant legal precedent for AI employment discrimination liability.

Research revealed that AI essay grading systems systematically gave lower scores to essays when student names suggested African American identity, demonstrating concerning racial bias in educational AI tools.

Google's Gemini AI image generator exhibited severe bias by refusing to create images of white people and generating historically inaccurate depictions. Google paused the feature after widespread criticism.

Google's Gemini AI refused to state that Hitler had a more negative impact than Elon Musk, citing policies against making comparisons between people. The incident highlighted problems with overly cautious AI safety guardrails that prevented reasonable moral judgments about historical figures.

Google's Gemini AI image generator refused to create images of white people and produced historically inaccurate diverse images due to overcorrected diversity guidelines, forcing Google to pause the feature amid widespread criticism.

Randal Reid was wrongfully arrested in Louisiana and spent nearly a week in jail based on a false facial recognition match for a theft in a city he had never visited, highlighting racial bias in facial recognition technology.

The FTC banned Rite Aid from using facial recognition for 5 years after the technology falsely identified customers as shoplifters, disproportionately harming Black, Latino, and Asian customers through wrongful detentions and searches.

FTC banned Rite Aid from using facial recognition for 5 years after the technology falsely identified customers as shoplifters, with disproportionate impact on minority customers.

UnitedHealthcare used an AI algorithm called nH Predict that systematically denied elderly patients' post-acute care coverage despite having a 90% error rate when reviewed by humans. A class action lawsuit was filed in 2023 alleging the practice harmed thousands of Medicare Advantage patients.

UnitedHealth used an AI algorithm with a known 90% error rate to deny post-acute care coverage for elderly Medicare Advantage patients, overriding physician recommendations and forcing premature discharges from nursing facilities.

Bloomberg research revealed OpenAI's DALL-E consistently generated lighter-skinned, male professionals while depicting service workers as darker-skinned, demonstrating systematic racial and gender bias in AI image generation.

LSEG's World-Check screening database used AI algorithms that falsely flagged hundreds of innocent people as terrorists or criminals, causing them to be denied banking services and face reputational harm.