EPS
← All batches·2605.16312

When Actions Disappear: Adversarial Action Removal in Self-Play Reinforcement Learning

topic: current_projecttop score: 80released: 2026-05-19first surfaced: 2026-05-19arXivPDFthreats2026-05-19

Authors: Arahan Kujur

arXiv · PDF

Summary

arXiv:2605. 16312v1 Announce Type: new Abstract: We study adversarial action masking in self-play reinforcement learning: an attacker selectively removes legal actions from a victim's action set.

Relevance

Read next because When Actions Disappear: Adversarial Action Removal in Self-Play Reinforcement Learning overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)". Matching terms: under, line. Source: arxiv cs.LG (Machine Learning).

Threat model

Potential threat/caveat for clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)": this item discusses robustness, adversarial.

Abstract

arXiv:2605.16312v1 Announce Type: new Abstract: We study adversarial action masking in self-play reinforcement learning: an attacker selectively removes legal actions from a victim's action set. Unlike observation or action perturbations, removal eliminates decision options before the agent acts. Across poker games scaling from 6 to 5,531 information states and two non-poker domains, learned masking causes substantially more damage than random masking and learned perturbation baselines. The attack persists across Q-learning, PPO, NFSP, neural NFSP, and DQN victims; transfers across agents; is amplified by self-play; and shows no recovery under extended masked training. Mechanistically, the adversary targets high-value decision points, captured by reach-weighted contingent action capacity (CAC$_w$) and a value-weighted refinement CAC$_v$. These results identify action availability as a distinct robustness surface in self-play RL.