Newest papers

AgentPolicyForge: From Executable Environments to Policy Benchmarks and Adaptive Guards

Chao Shang, Wanyu Du, Yi Zhang, Sandesh Swamy · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.28989 · cited by 0 works (OpenAlex) · open-access PDF

Japanese

カスタマイズされたエージェントのデプロイは、ツール、リソース、および実行条件に対するポリシーを定義します。これらのポリシーをタスクエージェントのプロンプトで提供することは、違反を信頼できる方法で防止しません。警備員の訓練には、これらのポリシーに合致するまたは違反するツール呼び出しの例が必要です。我々は、実行可能な環境を使用してポリシーベンチャーを構築し、ポリシーガードをトレーニングするために実行可能なフレームワークを導入します。このフレームワークを使用して、我々は、トレーニングデータ生成のための698のシナリオと、人によってレビューされたテストセットのための254の別々のハードアウトシナリオ

Machine translation (Meta M2M100); the English original follows.

Abstract

Customized agent deployments define policies over tools, resources, and execution conditions. Providing these policies in the task agent's prompt does not reliably prevent violations. Training a guard requires examples of tool calls that comply with or violate those policies. We introduce AgentPolicyForge, a framework that uses executable environments to construct policy benchmarks and iteratively train policy guards. With this framework, we construct PolicyGym, a dataset across ten policy families using 698 scenarios for training data generation and 254 separate held-out scenarios for human-reviewed test sets. Tool calls are executed and their outcomes recorded to help determine whether they comply with the policy. Every violating case uses a permitted tool, requiring checks of arguments, resources, and context rather than tool identity alone. The case construction process supports iterative guard training: an attacker generates requests using retrieved strategies, and the guard is fine-tuned on balanced, environment-verified examples. Training improves policy enforcement even with a fixed archive. Updating the archive further reduces in-domain attack success compared with keeping it fixed. We compare policy-aware task agent self-regulation with a trained policy guard. On PolicyGym, the final guard-based configuration achieves a 40-44% relative reduction in attack success rate and fewer benign rejections than policy prompting alone.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Executable, Guard (computer science), Computer science, Construct (python library), Enforcement, Task (project management)

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange