Our official repo for
AI can be surprisingly creative, outsmarting the researchers using and developing it. That can lead to scientific breakthroughs, superhuman capabilities, and generating new knowledge, but such creativity can also be mischievous, raising safety concerns. We crowd-sourced anecdotes from the AI community of when researchers were surprised by how creative, innovative, and/or mischievous AI was in their experiments. The result: 26 entertaining stories of AI outwitting humans, whether researchers or opponents.
Either
- Submit a PR directly here under the contributions folder
- Write up and email your story to me (aadharna@gmail.com), and I can add it
When writing up your anecdote, please note that we would like it to be understandable by as large an audience as possible while keeping technical rigour.
The story should give just enough background, set up what you expected to happen, what actually happened, and explain why it was surprising. If you like, you can even give your takeaway from the story!
Finally, don't feel compelled to match my use of .tex files. Write up the story in whatever readable form is easiest for you!
In the future, we would like this to be a self-sustaining collection managed by the community, and so are looking into additional options like creating a Wikipedia page.
citation:
@misc{dharna2026aifindsway,
title={AI Finds A Way},
author={Aaron Dharna and Cong Lu and Ryan Sullivan and Joel Lehman and Victoria Krakovna and Jeff Clune},
year={2026},
eprint={2608.23875},
archivePrefix={arXiv},
primaryClass={cs.AI},
url={https://arxiv.org/abs/2608.23875},
}