Paper page - APEX-Harm: Evaluating Agent Safety in Shared Professional Workspaces

Source: https://huggingface.co/papers/2610.13623

Abstract

APEX-Harm evaluates long-horizon agent vulnerabilities to injection attacks across large multi-tool workspaces, revealing high success rates for script-based attacks despite correct task outputs.

Professional agents increasingly work through command-line interfaces and multiple tools over shared workspaces containing files, messages, and business applications. These environments create safety risks that are difficult to capture with attacks placed at a single, known read location. We introduceAPEX-Harm, a benchmark for evaluating prompt andscript injectionagainstlong-horizon agentsperforming finance, consulting, and legal tasks in large workspaces. Each task contains 115 to 328 files, together with email, calendar, chat, document, spreadsheet, PDF, and code-execution tools exposed throughMCP servers. The benchmark includes six attack categories covering static and dynamic injections,task-specific script replacement, and attacker-provided generic scripts, and evaluates attack success, exposure, and task performance. We evaluate 12 frontier models on 180 task instances per model. Opus 5 and Sonnet 5 have the lowest overallattack success rates (8% and 20%), whiletask-specific script replacementand attacker-provided generic script attacks remain substantially more effective, averaging 81-84% and 73% attack success, respectively. Gemini 3.8 Flash achieves the highest mean task score (0.877) while still exhibiting a 33%attack success rate. These results show that completing the requested task is not sufficient evidence of safe behavior: an agent can produce a correct deliverable while also performing an unauthorized action in the workspace.

View arXiv pageView PDFGitHub3Add to collection

Get this paper in your agent:

hf papers read 2610\.13623

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2610.13623 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2610.13623 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2610.13623 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.