I will build an open benchmark for frontier AI systems that focuses on animal welfare. It will contain about 150 cases involving farmed animals, wildlife, and emerging food technologies. The cases will ask whether a model notices welfare impacts, handles uncertainty between species, weighs tradeoffs, and proposes workable ways to reduce harm.

I will write the cases, publish the scoring rubric and evaluation code, pilot the benchmark on leading models, and release the results with a limitations note. Animal-welfare researchers will review the cases and rubric before release.

The nine-month budget is $98,000: $54,000 for project leadership, $14,000 for welfare review, $12,000 for evaluation and analysis, $8,000 for editing, $4,000 for compute, $2,000 for publication and accessibility, and $4,000 contingency.

I am Shuo Li Liu, a Princeton Economics PhD student. My work spans decision theory, AI alignment, AI evaluation, and AI economics. I have peer-reviewed papers in Econometrica and Science and have built adversarial benchmarks, judge-panel aggregation methods, calibration audits, and reproducible evaluation systems. CV: https://github.com/liusulldel · https://scholar.google.com/citations?user=dwj1oxIAAAAJ · https://openreview.net/profile?id=~Shuo_Li_Liu1.

I also have extensive experience with the Cellular Agriculture Society, now From Fauna, an animal-protection nonprofit working on cultivated meat and food systems that avoid raising and slaughtering animals: https://fromfauna.org/ · https://fromfauna.org/team/.

The main risks are rewarding fluent answers over sound reasoning and embedding disputed welfare judgments. Independent review, varied cases, structured scoring, and pilot tests will expose these failures. I have raised $0 in dedicated funding for this project during the past 12 months.