Research map / Adversarial robustness and security / Backdoor and poisoning
SleeperNets: Universal Backdoor Poisoning Attacks Against Reinforcement Learning Agents
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes SleeperNets, a universal backdoor poisoning attack against RL agents that poisons training with a dynamic reward-based trigger scheme, evaluated across RL environments and algorithms.
Problem
Training-time backdoor poisoning of reinforcement learning agents, where an adversary intercepts training to implant a trigger-conditioned malicious policy while keeping clean performance high.
Approach
The paper introduces SleeperNets, a universal poisoning scheme for RL that alters rewards according to a dynamic, time-varying function of the environment state and the injected trigger, aiming to be stealthy and effective across environments and algorithms. The abstract is truncated, so specifics beyond this are not confirmed from the provided text.
Closest papers by meaning
- BadRL: Sparse Targeted Backdoor Attack against Reinforcement Learning (AAAI 2024 · similarity 0.92)
- Adversarial Inception Backdoor Attacks against Reinforcement Learning (ICML 2025 · similarity 0.92)
- SHINE: Shielding Backdoors in Deep Reinforcement Learning (ICML 2024 · similarity 0.90)
- PolicyGuard: Towards Test-time and Step-level Adversary Defense for Reinforcement Learning Agent (ICML 2026 · similarity 0.89)
- Universal Jailbreak Backdoors from Poisoned Human Feedback (ICLR 2024 · similarity 0.88)
- Demystifying Poisoning Backdoor Attacks from a Statistical Perspective (ICLR 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.