The Hot Mess of AI: How Does Misalignment Scale With Model Intelligence and Task Complexity?

Record

Authors
Hägele, A.; Gema, A. P.; Sleight, H.; Perez, E.; Sohl-Dickstein, J.
Year
2026
Title
The Hot Mess of AI: How Does Misalignment Scale With Model Intelligence and Task Complexity?
Published in
arXiv
Type
Preprint
DOI
10.48550/arxiv.2601.23045
arXiv
arXiv:2601.23045
Links
https://alignment.anthropic.com/2026/hot-mess-of-ai/
https://arxiv.org/abs/2601.23045
https://github.com/haeggee/hot-mess-of-ai
Our note
The Hot Mess of AI: How Does Misalignment Scale with Model Intelligence and Task Complexity?
Verification
verified: its identifier resolves to a record with the same title and author (checked against OpenAlex, 2026-11-14).
Identifiers found: OpenAlex W7127176850 · DOI 10.48550/arxiv.2601.23045 · arXiv:2601.23045
From the lookup
doi (the pages' own wording is under "As cited" below)
Id
hagele-2026-hot-mess-ai-how-does-misalignment

Cited on

As cited

The pages cite this work in 2 forms.

BibLaTeX

@misc{hagele-2026-hot-mess-ai-how-does-misalignment,
  author = {Hägele, A. and Gema, A. P. and Sleight, H. and Perez, E. and Sohl-Dickstein, J.},
  title = {{The Hot Mess of AI: How Does Misalignment Scale With Model Intelligence and Task Complexity?}},
  date = {2026},
  doi = {10.48550/arxiv.2601.23045},
  eprint = {2601.23045},
  eprinttype = {arxiv},
  url = {https://alignment.anthropic.com/2026/hot-mess-of-ai/}
}

Download · JSON with CSL-JSON · the whole bibliography