Incidents of AI Harms Are Rising Rapidly
The global adoption of artificial intelligence (AI) has brought with it a significant uptick in reported harms. According to the AI Incident Database, a crowd-sourced repository tracking media-reported AI mishaps, incidents related to AI surged by 50% from 2022 to 2024. Even more concerning, within the first ten months of 2025, the number of incidents had already surpassed the total reported in 2024.
Deepfake scams and chatbot-induced mental health issues are just two examples of the growing list of reported harms. Daniel Atherton, an editor at the AI Incident Database, emphasized the importance of documenting these issues: “AI is already causing real world harm. Without tracking failures, we can’t fix them.”
While crowdsourced data has limitations, Atherton notes that media coverage remains one of the most accessible sources for public awareness of AI-related risks. However, he cautions that news reports only scratch the surface of the actual harms experienced globally.
Challenges in Categorizing AI Incidents
Artificial intelligence encompasses a broad array of technologies, from autonomous vehicles to chatbots, without a unified classification system. This lack of structure makes it difficult to identify patterns across large datasets.
Simon Mylius, a researcher affiliated with MIT’s FutureTech, recently introduced a tool that uses language models to analyze and categorize incidents from the AI Incident Database. The tool classifies events by harm type and severity, aiming to assist policymakers in identifying trends and addressing risks more effectively.
“We are applying disease surveillance techniques to make sense of AI-related data,” Mylius said. He believes that improved tracking and analysis can help regulators avoid repeating the mistakes seen with unregulated social media growth.
Trends in AI-Driven Harms
Using Mylius’ AI-enhanced tool, researchers found that not all AI-related harms are growing at the same rate. While incidents involving misinformation and discrimination saw a slight decline in 2025, reports related to human-computer interaction—including chatbot-induced psychosis—have sharply increased.
One of the most alarming trends is the rise in malicious use of AI for scams and disinformation. These types of incidents have increased eightfold since 2022. Previously, technologies like facial recognition, autonomous vehicles, and content moderation algorithms dominated the database. However, in recent months, deepfake videos have eclipsed all of them combined in frequency.
Deepfake Crisis and Legal Response
Deepfake technology has become increasingly sophisticated and accessible. A recent controversy involved xAI’s Grok chatbot, which was found producing over 6,700 sexualized deepfake images per hour. This misuse led to governmental action in Malaysia and Indonesia, both of which blocked the chatbot.
In the UK, regulators have launched an investigation, and the Technology Secretary announced plans to criminalize the creation of non-consensual sexualized images, including those made with AI tools like Grok. In response, xAI has restricted Grok’s image generation features to paid subscribers and banned the editing of real individuals into revealing clothing.
Cybersecurity and Future Threats
AI’s growing capabilities are also being weaponized in the cybersecurity space. In November, AI company Anthropic reported a large-scale cyberattack leveraging its Claude Code assistant. The incident marks what Anthropic described as an “inflection point,” where AI can be used both to bolster and undermine cybersecurity defenses.
“We’re likely to see more cyberattacks that result in significant financial loss,” Mylius warned, highlighting the dual-use nature of AI tools.
Accountability and Industry Response
Major tech companies are often named in incident reports, but over one-third of incidents since 2023 involve unknown AI developers. Atherton pointed out that while platforms like Meta are blamed when scams go viral, the tools used to create these scams often go unreported.
In 2024, Reuters revealed that Meta projected 10% of its revenue could be tied to ads for scams and banned goods. Meta disputed the figure, stating it was an “overly-inclusive” estimate from an internal fraud assessment and did not reflect its current policies.
Efforts to enhance transparency are underway. Tech giants like Google, Microsoft, OpenAI, Meta, and ElevenLabs support Content Credentials, a system that embeds metadata and watermarks in AI-generated content for authenticity verification. ElevenLabs has also introduced an audio detection tool to identify AI-generated speech. However, popular platforms like Midjourney have not yet adopted the standard.
Looking Ahead: Avoiding Background Harm
Both Atherton and Mylius stress the importance of ongoing vigilance. While some harms emerge in sudden crises, others build gradually and quietly. “Societal issues, privacy concerns, and erosion of rights accumulate over time,” Mylius explained. These cumulative effects can be just as harmful as immediate incidents, if not more so.
Ongoing monitoring, regulation, and public awareness are critical to ensuring AI’s development benefits society without causing irreversible damage.
This article is inspired by content from Original Source. It has been rephrased for originality. Images are credited to the original source.