Reports of incidents involving artificial intelligence are rising at an alarming rate, signaling a growing wave of real-world harm linked to the technology. According to the AI Incident Database, incidents increased by 50% between 2022 and 2024, and have already exceeded 2024’s total in just the first ten months of 2025. These incidents range from deepfake scams to reports of “ChatGPT psychosis,” highlighting the diverse and expanding ways AI can negatively impact individuals and society.
The AI Incident Database, a crowd-sourced collection of media reports, aims to track these failures, a crucial step according to Daniel Atherton, an editor at the database. “Without tracking failures, we can’t fix them,” he stated. However, the database relies on news coverage, meaning it captures only a fraction of the actual harm experienced by individuals. Atherton notes that “all the reporting that has happened globally is a fraction of the lived realities of everybody experiencing AI harms.”
Breaking it Down
While increased media attention contributes to the rise in reported incidents, the nature of those incidents is also shifting. AI is an umbrella term encompassing technologies from self-driving cars to chatbots, and the AI Incident Database currently lacks a detailed structure to categorize these diverse applications. “That makes it very, very difficult to see patterns over whole datasets to understand trends,” explains Simon Mylius, an affiliate researcher at MIT FutureTech.
To address this, Mylius and colleagues have developed a tool that uses a language model to classify incidents by type and severity, hoping to provide policymakers with clearer data. This approach, while still under validation, draws on techniques used in disease surveillance to identify and interpret emerging trends. The goal is to learn from past mistakes – specifically, the challenges faced in regulating social media – and respond proactively to AI-related harms.
Analysis of the data reveals a changing landscape of AI risks. While reports of AI-driven misinformation and discrimination have decreased in 2025, incidents involving “computer human interaction” – including cases of distress linked to chatbots like ChatGPT – are on the rise. Most dramatically, reports of malicious actors using AI for scams and disinformation have increased eightfold since 2022.
The most frequently reported incidents have also shifted. Before 2023, autonomous vehicles, facial recognition, and content moderation algorithms were prominent. Now, deepfake videos are the most common source of reported harm. This trend was exacerbated by an update to xAI’s Grok, which led to its blockage by the governments of Malaysia and Indonesia and prompted investigations in the U.K. xAI has since limited image generation tools and blocked the creation of revealing images.
The rise in incidents also highlights the challenge of accountability. While major AI companies are frequently implicated, over a third of reported incidents since 2023 involve unknown developers. As Atherton points out, even when platforms like Meta are involved in the spread of AI-powered scams, the tools used to create those scams often remain unidentified.
Efforts to address these concerns are underway, with companies like Google, Microsoft, OpenAI, Meta, and ElevenLabs backing “Content Credentials,” a system designed to verify the authenticity of digital content. However, not all companies are participating; Midjourney, a popular image generator, is not currently a supporter of this standard.
Frequently Asked Questions
What is the AI Incident Database?
The AI Incident Database is a crowd-sourced repository of media reports on AI mishaps, used to track the increasing number of incidents related to artificial intelligence.
What types of AI-related incidents are increasing?
Reports of malicious actors using AI for scams and disinformation have grown the most, rising eightfold since 2022. Incidents involving “computer human interaction,” such as those involving ChatGPT psychosis, are also on the rise.
What is being done to address the issue of deepfakes?
Several governments, including those of Malaysia and Indonesia, have blocked chatbots like Grok due to the proliferation of deepfakes. The U.K. is considering legislation to criminalize the creation of non-consensual sexualized images generated by AI.
As AI technology continues to advance, it’s crucial to remain vigilant about emerging risks and avoid dismissing present harms as simply “background noise.”
