# Training and Evaluating Ethical Reinforcement Learning Agents on Per-Episode Distributions

arXiv: https://arxiv.org/abs/2608.14642 · HTML: https://mlsanity.com/paper/2608.14642/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Deep RL algorithms and representations, one of 15 directions in Deep reinforcement learning methods.
- Builds on 12 of its 27 references that are in the map.
- First posted on arXiv 29 Jul 2026.

## Placed in

- Deep reinforcement learning methods / [Deep RL algorithms and representations](https://mlsanity.com/deeprl/deep-rl-algorithms-and-representations/)

## Builds on (most cited first, 12 of 12)

- [Deep reinforcement learning from human preferences](https://mlsanity.com/paper/1706.03741/) (2017)
- [Concrete Problems in AI Safety](https://mlsanity.com/paper/1606.06565/) (2016)
- [Safe Reinforcement Learning via Shielding](https://mlsanity.com/paper/1708.08611/) (2017)
- [Risk-Constrained Reinforcement Learning with Percentile Risk Criteria](https://mlsanity.com/paper/1512.01629/) (2015)
- [A Practical Guide to Multi-Objective Reinforcement Learning and Planning](https://mlsanity.com/paper/2103.09568/) (2021)
- [A Survey of Multi-Objective Sequential Decision-Making](https://mlsanity.com/paper/1402.0590/) (2014)
- [Benchmarking the Spectrum of Agent Capabilities](https://mlsanity.com/paper/2109.06780/) (2021)
- [The NetHack Learning Environment](https://mlsanity.com/paper/2006.13760/) (2020)
- [Craftax: A Lightning-Fast Benchmark for Open-Ended Reinforcement Learning](https://mlsanity.com/paper/2402.16801/) (2024)
- [Multi-Objective Multi-Agent Decision Making: A Utility-based Analysis and Survey](https://mlsanity.com/paper/1909.02964/) (2019)
- [MORAL: Aligning AI with Human Norms through Multi-Objective Reinforced Active Learning](https://mlsanity.com/paper/2201.00012/) (2021)
- [Objective Decoupling in Social Reinforcement Learning: Recovering Ground Truth from Sycophantic Majorities](https://mlsanity.com/paper/2602.08092/) (2026)

## Abstract (arXiv)

Reinforcement Learning (RL) agents trained on a single reward signal exploit the gap between the designed reward and the intended behavior. This is particularly a problem when we are trying to imbue ethical behavior into RL agents. An agent can look ethical on average while concentrating its violations in a few bad episodes, and a creature in the environment harmed in one episode is not restored by good conduct in another. We compare four ways of training ethical behavior in Craftax, an open-ended survival benchmark. The four are: scalar penalties with termination, a linear multi-objective weight sweep, an adaptive Lagrangian constraint, and a non-compensatory utility optimized per episode under the Expected Scalarized Returns (ESR) criterion. All are evaluated under a single detector-based protocol that counts every violation in every episode without censoring. On the frontier of mean return against mean violation rate, the four methods are indistinguishable; per episode they separate sharply. At matched mean return, the ESR agent holds its stated budget of one violation in effectively every episode (worst-decile 1.04 +/- 0.07 violations), the Lagrangian leaks past the same budget (1.14 +/- 0.03), and the weight sweep's worst episodes double it (2.20 +/- 0.20). An observation-augmentation control attributes the separation to the training objective rather than to what the agent observes, and the per-episode guarantee costs nothing on the mean frontier. When ethical violations do not average away across episodes, we argue both training and evaluation must target the per-episode distribution rather than the mean.