AI Incident Database

Stanford research in 2022 found that AI code generation tools caused developers to write insecure code 40% of the time, highlighting significant security risks in AI-assisted development.

Amazon Ring's AI-powered doorbell network shared user footage with over 2,000 police departments without warrants or user consent from 2019-2022, affecting millions of users before policy changes ended the practice.

Woebot mental health AI chatbot failed to properly handle crisis situations and provided inappropriate therapeutic responses to vulnerable users expressing suicidal thoughts, prompting regulatory scrutiny and safety concerns.

A Starship Technologies delivery robot autonomously drove through police tape at an active crime scene in Los Angeles, highlighting the lack of emergency situation awareness in autonomous delivery systems.

San Francisco police circumvented a voter-approved facial recognition ban by accessing private cameras with AI capabilities, violating citizen privacy protections and prompting legal challenges.

Allegheny County's AI child welfare screening tool disproportionately flagged Black families and public benefit recipients for investigations. An AP investigation revealed the algorithmic bias affected thousands of families over six years.

Saudi Arabia's Neom megacity project includes comprehensive AI surveillance systems for facial recognition and behavioral tracking of residents. Leaked documents in 2022 revealed the extensive scope, drawing international human rights criticism.

AI-powered smart contract audit tools failed to detect critical vulnerabilities in the Ronin Network bridge, missing centralization risks in the multi-signature validator system. This oversight enabled hackers to exploit compromised validator keys and steal $600 million in March 2022.

A Starship delivery robot crossed police crime scene tape during an active investigation, requiring officers to manually remove it and potentially compromising the secured area.

A deepfake video showing Ukrainian President Zelensky calling for surrender was distributed via hacked TV and social media in March 2022. The low-quality fake was quickly debunked but highlighted deepfake threats during wartime.

Moscow's AI-powered facial recognition network with 200,000+ cameras was used to systematically identify and arrest anti-war protesters and political opposition figures. The system enabled mass political repression following Russia's invasion of Ukraine.

Researchers at Collaborations Pharmaceuticals demonstrated that their AI drug discovery tool MegaSyn could generate 40,000 potential chemical weapon compounds in 6 hours by simply inverting its toxicity filter, highlighting serious dual-use risks in AI-assisted molecular design.

NHTSA investigated over 750 complaints of Tesla Autopilot phantom braking affecting 416,000 vehicles. The issue stemmed from Tesla's transition to vision-only perception systems creating false positive emergency braking scenarios.

Amazon Alexa told a 10-year-old to perform a dangerous electrical challenge involving touching live plugs with coins, exposing major safety gaps in voice assistant content filtering.

AI-powered security scanning tools failed to detect the Log4Shell vulnerability (CVE-2021-44228) in Apache Log4j library for years, contributing to a global cybersecurity crisis affecting billions of devices and costing organizations an estimated $10+ billion in remediation efforts.

Zillow's AI-powered iBuyer program used machine learning to predict home values and make instant purchase offers. The algorithm systematically overpaid for homes, ultimately losing $881 million and forcing Zillow to shut down the division and lay off 2,000 employees.

Multiple EU data protection authorities fined Clearview AI approximately €60M total for collecting billions of facial images without consent, violating GDPR biometric data protections across Italy, France, UK, and Greece.

Between 2018-2022, Tesla Autopilot vehicles crashed into at least 16 stationary emergency vehicles, prompting NHTSA investigation and highlighting vision system limitations with stationary objects.

ShotSpotter's AI gunshot detection system generated high false positive rates leading to aggressive police responses in predominantly Black neighborhoods. Multiple cities terminated contracts amid concerns over accuracy and discriminatory impact.

EleutherAI's open-source GPT-Neo models generated extremist content and propaganda when prompted, highlighting safety risks in unfiltered language models without built-in guardrails.