Reports of AI misuse have grown sharply, according to the Artificial Intelligence Index Report 2023. The rise points both to AI’s expanding presence in everyday settings and to increased attention to the harms its use can cause. But improved reporting also means older cases may have been missed.
More incidents are being recorded
The Stanford University report, now in its sixth year, devotes a chapter to technical AI ethics. The 2017 edition did not mention “deepfake”; the 2023 report mentions it 21 times. That change reflects how the discussion around AI has broadened.
The independent AIAAIC database, which tracks incidents involving AI misuse, counted a 26-fold increase between 2012 and 2021. That figure describes recorded cases. It does not, by itself, show how much of the change came from more misuse and how much came from better detection and documentation.
Final figures for 2022 were not available because database entries undergo a lengthy verification process. The report also cautions that many cases may have gone unreported in the database’s early years. As tracking improves, comparisons across time need to account for that gap.
Deepfakes can carry a message far
The report points to a deepfake video published in March 2022 as an example of an incident with particularly high reach. In it, Ukrainian President Volodymyr Zelenskyy allegedly calls on his troops to surrender.
The example shows how synthetic media can put words into a public figure’s mouth and spread a false message. The report’s broader concern is not limited to this one video: the growing visibility of deepfakes has brought questions about misuse into sharper focus.
AI in surveillance raises fairness questions
AI use in monitoring has also prompted ethical debate. The article describes reports of AI systems being used to monitor phone calls in U.S. prisons. Speech recognition was found to produce less accurate transcripts for Black individuals, even though a large proportion of prisoners in the U.S. are Black.
That example links two issues: surveillance can affect people in constrained settings, and the systems used may perform unevenly across groups. When automated transcripts inform how a conversation is understood, differences in accuracy can matter.
Monitoring is not always hidden. Intel has developed software for the videoconferencing tool Zoom that can read students’ facial expressions as “bored,” “distracted,” or “confused.” Such labels invite questions about how reliably a system can interpret emotion from a face and how that information might be used.
Image generators widen the debate
The report also raises concerns about AI image generators, including Midjourney, DALL-E 2 and Stable Diffusion. The issues it names include fairness, bias, copyright, jobs and privacy. These questions apply across the tools, rather than being confined to a single product.
The article notes that the report does not discuss the ability to generate photorealistic images, a capability that had recently attracted attention in connection with Pope Francis and Donald Trump. The omission highlights how quickly the public conversation around generative AI can expand beyond the concerns already covered in a report.
Together, these examples show why incident counts matter, while also showing their limits. More recorded cases can signal a wider footprint for AI and stronger awareness of potential harms; at the same time, earlier underreporting makes the history incomplete. Tracking and ethical scrutiny are both part of understanding how AI is being used.