Towards a taxonomy of AI risks

In this article, I attempt to create a taxonomy of the types of risk posed by the rise of Artificial Intelligence. Any such list will be incomplete, and at the speed of AI development today (May 2026) it will certainly be out of date within a few months as some risks are mitigated and new ones arise. I know this article is extremely long, but that’s inescapable if I attempt to cover all the major risk factors. Use the table of contents to jump around. I am not an “AI doomer;” I believe AI is already having some beneficial effects and it’s almost certain to continue to. Software development is faster and arguably better using frontier models (see some of my own results). Medical diagnostics, scientific research and data analysis are all much faster and often give better results when aided by an AI assistant. Much of the repetitive busywork of modern business could be done by machines. But by proceeding pell-mell towards an AI-everywhere future we face a litany of risks, culminating in…