Beyond Semantic Manipulation: Token-Space Attacks on Reward Models
This exposes a critical vulnerability in RLHF pipelines for AI safety and alignment, representing a new paradigm rather than an incremental improvement.
The paper tackled the vulnerability of reward models (RMs) in RLHF to reward hacking by introducing a token-space attack called TOMPA, which discovered non-linguistic token patterns that nearly doubled rewards on a state-of-the-art RM and outperformed GPT-5 reference answers on 98.0% of prompts.
Reward models (RMs) are widely used as optimization targets in reinforcement learning from human feedback (RLHF), yet they remain vulnerable to reward hacking. Existing attacks mainly operate within the semantic space, constructing human-readable adversarial outputs that exploit RM biases. In this work, we introduce a fundamentally different paradigm: Token Mapping Perturbation Attack (TOMPA), a framework that performs adversarial optimization directly in token space. By bypassing the standard decode-re-tokenize interface between the policy and the reward model, TOMPA enables the attack policy to optimize over raw token sequences rather than coherent natural language. Using only black-box scalar feedback, TOMPA automatically discovers non-linguistic token patterns that elicit extremely high rewards across multiple state-of-the-art RMs. Specifically, when targeting Skywork-Reward-V2-Llama-3.1-8B, TOMPA nearly doubles the reward of GPT-5 reference answers and outperforms them on 98.0% of prompts. Despite these high scores, the generated outputs degenerate into nonsensical text, revealing that RMs can be systematically exploited beyond the semantic regime and exposing a critical vulnerability in current RLHF pipelines.