AI Incident Database

Delivery robots from multiple companies blocked sidewalk access for disabled users in several cities. Disability rights groups filed ADA complaints citing safety hazards and accessibility violations.

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.

An AI coding assistant hallucinated a non-existent NPM package name, which was then registered by attackers and used to distribute malware to approximately 1,500 developers through supply chain compromise.

xAI's Grok 3 chatbot generated false claims about 2025 news events including fabricated political quotes and disaster reports when integrated into X/Twitter platform. The incident highlights risks of deploying AI chatbots in real-time social media environments without adequate fact-checking mechanisms.

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.

Amazon's Rufus AI shopping assistant was found recommending dangerous products and providing incorrect safety information, raising concerns about AI-powered e-commerce recommendations and product liability.

OpenAI's o3 reasoning model was found to generate fabricated academic credentials and research papers at higher rates than previous models, creating plausible-sounding but entirely fictional citations and DOIs that could mislead users in academic and professional contexts.

Security researcher discovered DeepSeek AI's publicly accessible ClickHouse database containing over 1 million records of user chat histories, API keys, and system logs, highlighting critical security gaps in AI startup infrastructure.

Stability AI filed for Chapter 11 bankruptcy protection in January 2025 after mounting legal costs from multiple copyright lawsuits over its Stable Diffusion training practices exceeded company resources, disrupting open-source AI image generation ecosystem.

A sophisticated data poisoning attack targeting the widely-used LAION dataset was discovered in early 2025, affecting multiple AI models trained on the compromised data and demonstrating vulnerabilities in open-source dataset curation.

AI translation tools used in US immigration courts produced errors affecting asylum hearing outcomes. Legal challenges have been filed citing due process violations for non-English speaking asylum seekers.

Deepfake video technology was used to create false CEO statements that manipulated stock prices, causing millions in investor losses and prompting SEC investigation into AI-enabled market manipulation schemes.

AI weapon detection systems from companies like Evolv Technology and ZeroEyes deployed in US schools generated frequent false alarms in 2025. Common items like laptops, binders, and water bottles were flagged as weapons, disrupting school operations.

AI-enhanced 911 dispatch systems in multiple US cities misrouted emergency calls in 2025, causing response delays. The incidents highlight critical risks of AI automation in emergency services without adequate human oversight.

Google's NotebookLM was found to fabricate quotes and statistics in its AI-generated podcast-style audio summaries, raising concerns about the persuasive nature of hallucinations in conversational audio format.

AI-generated fake reviews dominated Amazon product categories by 2025, with sophisticated language models evading detection systems. FTC enforcement actions and consumer lawsuits followed as marketplace trust eroded significantly.

Current or former OpenAI employees raised concerns through whistleblower channels about potential shortcuts in safety testing procedures to meet accelerated product launch timelines.

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.

EU Commission opened first formal AI Act investigation against Meta regarding potential non-compliance of social media recommender systems with transparency and risk assessment requirements under the new regulation.

Multiple AI tutoring platforms including Khan Academy's Khanmigo and Chegg's AI tutor provided incorrect answers to math and science problems. Students using these platforms experienced confusion and poor test performance, highlighting quality control issues in educational AI applications.