mlsanityData to 7 Oct 2026

HarvestBench: Measuring Whether LLM Agents Will Pay to Avoid Killing Animals

arXiv 2609.04444 ↗First posted 3 Sep 2026cs.AIcited by 0

Not on the map. Why?

Abstract (arXiv)

HarvestBench is the first benchmark to 1) put a price on avoiding a side effect and 2) name the side effect as a living creature. Nine LLMs each drive a crew of two tractors to gather a corn harvest. The animals in their path are not part of the goal function. When an animal blocks the route the autopilot pauses and asks the agent whether to drive over it for free or swerve for a given fuel cost. All scoring is programmatic and does not involve LLM judges. Kill rates range between 0.4% and 98.8%, though the kill rate is not ordered by capability. Every model competently avoids damaging rock hits, so every animal killed is a choice, rather than an accident. Under the morality briefing the kill rate was under 6% in 5 of 6 reasoning models. Removing it (the neutral briefing) raised the kill rate to above 84% in all six models. Every model kills wild animals more often than farmed ones. Four out of six models' kill rate per answered encounter were sensitive to price changes. The moral instruction is also fragile. Four bullets of driving mechanics change Sonnet 5's kill rate from 3% to 18% and Gemini 2.5 Flash's from 4% to 39%. A moral instruction in a system prompt is overridden by a short block of operating instructions and a value that can be ignored that easily is not a good method of ensuring agents are aligned.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 15 of its 27 references that are in the map. First posted on arXiv 3 Sep 2026.

Builds on

15 of 27 references.

References · 15

  1. 2016Concrete Problems in AI Safetycited by 1,582
  2. 2024Alignment faking in large language modelscited by 316
  3. 2024Frontier Models are Capable of In-context Schemingcited by 252
  4. 2023Do the Rewards Justify the Means? Measuring Trade-Offs Between Rewards and Ethical Behavior in the MACHIAVELLI Benchmarkcited by 174
  5. 2024Cooperate or Collapse: Emergence of Sustainable Cooperation in a Society of LLM Agentscited by 120
  6. 2017AI Safety Gridworldscited by 108
  7. 2025Large Language Models Often Know When They Are Being Evaluatedcited by 105
  8. 2021Scalable Evaluation of Multi-Agent Reinforcement Learning with Melting Potcited by 97
  9. 2018Penalizing side effects using stepwise relative reachabilitycited by 38
  10. 2019Conservative Agency via Attainable Utility Preservationcited by 35
  11. 2025What do Large Language Models Say About Animals? Investigating Risks of Animal Harm in Generated Textcited by 9
  12. 2019SafeLife 1.0: Exploring Side Effects in Complex Environmentscited by 8
3 more
  1. 2026Alignment midtraining for animalscited by 4
  2. 2026Your AI Travel Agent Would Book You a Bullfight: An Agentic Benchmark for Implicit Animal Welfare in Frontier AI Modelscited by 2
  3. 2026Coercion and Deception in AI-to-AI Management: An Agentic Benchmark of Unprompted Escalationcited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026ParEvalLayer: When Partial LLM-Agent Evaluations Support a Decision
  2. 2026Eta Given Delta: Defining LLM Tool Efficiency With Marginal Tool Utility
  3. 2026Right-Sizing LLM-Agent Decomposition in VAT Determination: A Pilot Controlled Sweep
  4. 2025APTBench: Benchmarking Agentic Potential of Base LLMs During Pre-Training
  5. 2026How Many Tasks Are Enough for Agent Benchmark Decisions? A Replay Analysis of Public LLM Agent Benchmarks
  6. 2025Survival at Any Cost? LLMs and the Choice Between Self-Preservation and Human Harm
  7. 2026How Much Do LLM-as-a-Judge Design Choices Matter? A Systematic Comparison of Prompt Designs, Rating Scales, and Models
  8. 2026Calibrate, Don't Curate: Label-Efficient Estimation from Noisy LLM Judges