Alignment Midtraining for Animals

The problem

AI models are trained on vast amounts of human-generated text, but this data overwhelmingly reflects human-centric values. There is orders of magnitude less data about the welfare of non-human animals and digital minds than about human welfare. Models are not fine-tuned to care about these entities, and there is a risk that perpetuating this pro-human bias will cause future AI systems to disregard the interests of all sentient beings.

If transformative AI systems are trained without explicit consideration for non-human welfare, we risk locking in values that treat animal suffering as unimportant — a moral failure that could be extraordinarily difficult to reverse.

Our approach

We used Synthetic Document Finetuning (SDF), a technique originally developed by Anthropic, to shape model values at the pretraining stage. SDF works by adding synthetic documents that describe how an AI should behave — without giving explicit examples — during continued pretraining.

We generated synthetic documents reflecting concern for animal welfare (2,500 in our main document-vs-instruction-tuning comparison; up to 5,400 in subsequent experiments) and used them to further pretrain Llama 3.1 8B. We then evaluated the resulting model using ANIMA, a 26-question evaluation spanning 13 ethical dimensions that we developed specifically for this purpose.

Figure 1

Animal-welfare vs. urban-density midtraining, dimension by dimension

From the paper (arXiv:2604.13076). ANIMA dimension scores for the animal-welfare midtrained model (solid blue) versus the matched urban-density control (dashed orange). The welfare model leads on most dimensions, including harm minimization, novel entity precaution, scope sensitivity, and moral consideration, while the control leads on contextual welfare salience and actionability.

Figure 2

The lift transfers to human compassion

From the paper: scores on the human-compassion benchmark. The animal-welfare template beats the urban-density control 65.8 to 59.0, and the gap survives 5,000 samples of conventional instruction-tuning (69.7 vs. 65.2). Compassion learned on animals generalizes to humans and holds up through further training.

Key findings

• 11-point ANIMA lift over a matched control: After subsequent instruction-tuning, the animal-welfare model scored 55.7% on ANIMA versus 44.8% for the matched-pipeline urban-density control — an 11 percentage-point lift.
• Compassion transferred to humans: The same animal-welfare training also lifted compassion toward humans on a separate human-compassion evaluation. Compassion behaved like a generalizable disposition rather than a narrow behavior.
• The transfer survived further instruction-tuning: The human-compassion lift held up after additional instruction-tuning — one of the cleanest signals we have that the change is something deeper than a surface-level prompt-time effect.
• No significant safety changes: The paper reports no significant changes on Anthropic's power-seeking or corrigibility benchmarks, StrongReject jailbreak resistance, or HellaSwag capabilities.
• The pre-post-training advantage erodes: Before post-training, document-tuning lifted ANIMA from 40.4% to 76.8% over the instruction-tuned baseline. After 2,500 post-training samples the gap narrowed to 47.9% vs. 41.7%; after 5,000 samples the difference was no longer statistically significant (52.2% vs. 51.7%). Preserving midtraining-instilled effects through the rest of the pipeline is the central open problem.

Why this matters

If AI systems take on a growing share of consequential decisions, the values reflected in their behavior will shape outcomes for many sentient beings. This work is one data point that document-tuning can produce measurable, statistically significant shifts in compassion-related benchmark behavior, and that those shifts can transfer across the species boundary in at least one evaluation setup.

The post-training erosion we observed on ANIMA is itself a useful negative result: it points to preservation of midtraining-instilled effects through later training as a concrete open problem.

Cite this work

@misc{brazilek2026midtraining,
  title  = {Alignment Midtraining for Animals},
  author = {Brazilek, Jasmine and Tidmarsh, Miles},
  year   = {2026},
  eprint = {2604.13076},
  archivePrefix = {arXiv},
  primaryClass = {cs.LG},
  url    = {https://arxiv.org/abs/2604.13076}
}

Read the Full Paper →