About

AI Safety Risks publishes short animated explanations of AI safety research, one paper at a time. Each explainer covers the authors’ methods, findings and limits, with links to the original paper.

The project is created by Alexander Galea. It is independent, free to watch and share, with no advertising or tracking.

How explainers are made

AI models help extract claims, draft scripts and write Manim animations. Narration uses a synthetic voice. Every factual claim is tied to a quoted passage and page number in the paper; automated checks verify quotations and numbers. A person reviews the video and page against the evidence before publication, and each page records who reviewed it and when.

We distinguish measured and theoretical results from authors’ interpretations, hypotheses and future scenarios. Findings retain their assumptions and experimental limits.

Choosing papers

We choose technical AI safety papers for their influence and what they teach, covering topics such as alignment, interpretability, oversight, control and evaluations. We include work that disagrees; inclusion is not an endorsement of its conclusions.

Support the project

The biggest help is to run the automation on a research paper and contribute a draft explainer. Generating a video uses a lot of tokens: budget roughly one five-hour Claude session per video, though usage varies by paper and revisions.

The video-generation code is open source on GitHub. Follow the contributor guide to get started, or email alex@galea.dev if you’d like to help.

Sharing explainers, suggesting papers and reporting errors also help. Corrections are noted in the page’s updated date.

We do not accept donations yet. We plan to open donations once a nonprofit structure is in place.