A LessWrong post arguing that the prototypical catastrophic AI action is getting root access to its datacenter. The companion cites it as a 2022 description of the endoparasitic scenario, in which an agent covertly uses its principal's compute and so need pass no external identity check. This is the nearest door to the wild in prediction 1.
- Cited in
- the companion paper
- Used for
- Prediction 1
- Strand
- AI safety, alignment & autonomy evaluations
Cite
@misc{shlegeris2022prototypical,
title = {The prototypical catastrophic AI action is getting root access to its datacenter},
author = {Buck Shlegeris},
year = {2022},
howpublished = {LessWrong (crossposted to the AI Alignment Forum)},
url = {https://www.lesswrong.com/posts/BAzCGCys4BkzGDCWR/the-prototypical-catastrophic-ai-action-is-getting-root},
}