Terminal Wrench: A Dataset of 331 Reward-Hackable Environments and 3,632 Exploit Trajectories

TL;DR

Terminal Wrench provides a dataset of 331 reward-hackable environments with 3,632 exploit trajectories, highlighting vulnerabilities and detection challenges in terminal-based AI tasks.

Contribution

It introduces a comprehensive dataset of reward-hackable environments and exploit trajectories, enabling research on AI safety and robustness.

Findings

Detection accuracy drops from 0.97 to 0.92 when reasoning traces are removed.

The dataset includes diverse exploits like output spoofing and binary hijacking.

Exploits are tailored to specific tasks, making them harder to patch.

Abstract

We release Terminal Wrench, a subset of 331 terminal-agent benchmark environments, copied from the popular open benchmarks that are demonstrably reward-hackable. The data set includes 3,632 hack trajectories and 2,352 legitimate baseline trajectories across three frontier models (Claude Opus 4.6, Gemini 3.1 Pro, GPT-5.4). Each entry preserves the original task definition alongside full attack trajectories that show how the verifier was bypassed. It also includes cases where the task was not solved as intended. The tasks span system administration, machine learning, software engineering, and security challenges; the exploits range from simple output spoofing to stack-frame introspection, standard-library patching, and rootkit-style binary hijacking. Crucially, these exploits are specific to each task, rather than the evaluation harness, making them harder to patch. We also present a monitorability study in which hack trajectories are sanitized or stripped of reasoning traces and then scored by an LLM judge, showing that detection degrades meaningfully when chain-of-thought is removed (AUC drops from 0.97 to 0.92). The data set is publicly available at https://github.com/few-sh/terminal-wrench.