Ten problems in AI right now
A sourced survey of ten areas of AI risk, with fixes and gaps for each.
I researched ten problem areas in AI as of October 2026. They include hallucination, prompt injection, runaway agent costs, energy use, privacy, bias, copyright, misuse, benchmark reliability, and running AI on your own servers.
Each item has a source, a known fix where one exists, and an honest note on what is unverified.
One example is fabricated references. About 1 in 277 biomedical papers in early 2026 had one, up from about 1 in 2,828 in 2023. Another is cost, since 79 percent of surveyed companies reported going over budget on AI.
It also lists open questions, because for most topics I did not go past the first level of detail.
