An adversarial testing framework for AI-driven task routing systems

Authors

DOI:

https://doi.org/10.31649/1681-7893-2026-51-1-374-381

Keywords:

Prompt injection, adversarial testing, task routing systems, large language models, enterprise information systems.

Abstract

The paper presents an adversarial testing methodology for evaluating AI-driven task routing systems. The methodology defines structured attack scenarios and strict output constraints to measure resistance against unauthorized data disclosure. To validate suggested  approach, an AI-based routing solution implemented using an Salesforce Agentforce Prompt Template powered by ChatGPT 5 was tested in a controlled environment. It has been proven that using a structured approach to testing can reduce the risk of data leakage in AI-based decision support systems.

Author Biographies

R.V. Slobodian, Vinnytsia National Technical University

PhD student of Automation and Intelligent Information Technologies Department

I.V. Bogach, Vinnytsia National Technical University

Associate Professor of Automation and Intelligent Information Technologies Department

References

The role of artificial intelligence in business transformation: A case of pharmaceutical companies [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0160791X21001044

Artificial intelligence as a driver of business process transformation [Online]. Available: https://www.sciencedirect.com/science/article/pii/S1877050922017586

Navigating the organizational AI journey: The AI transformation framework [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0007681325000023

Integrative innovation of large language models in industries: technologies, applications, and challenges [Online]. Available: https://www.sciencedirect.com/science/article/pii/S2666764925000323

Artificial Intelligence Applications [Online]. Available: https://www.sciencedirect.com/topics/computer-science/artificial-intelligence-applications

Data Inference: Data Security Threats in the AI Era [Online]. Available: https://www.sciencedirect.com/science/article/pii/S2095809925004722

A data-driven risk assessment of cybersecurity challenges posed by generative AI [Online]. Available: https://www.sciencedirect.com/science/article/pii/S2772662225000360

Data Leak: Meaning and Examples [Online]. Available: https://www.fortinet.com/resources/cyberglossary/data-leak

Risks and Consequences of Data Breach [Online]. Available: https://searchinform.com/articles/cybersecurity/cyber-threats/type/data-breach/risks-and-consequences/

What are large language models (LLMs)? [Online]. Available: https://www.ibm.com/think/topics/large-language-models

What are large language models (LLMs)? [Online]. Available: https://www.elastic.co/what-is/large-language-models

How do LLM security vulnerabilities differ from traditional application vulnerabilities? [Online]. Available: https://www.tencentcloud.com/techpedia/132460

Understanding prompt injections: a frontier security challenge [Online]. Available: https://openai.com/index/prompt-injections/

What Is a Prompt Injection Attack? [Examples & Prevention] [Online]. Available: https://www.paloaltonetworks.com/cyberpedia/what-is-a-prompt-injection-attack

Introducing the New Salesforce Developer Edition, Now with Agentforce and Data Cloud [Online]. Available: https://developer.salesforce.com/blogs/2025/03/introducing-the-new-salesforce-developer-edition-now-with-agentforce-and-data-cloud

Similarity-driven adversarial testing of neural networks [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0950705124012553

Best Practices for Secure Agentforce Implementation [Online]. Available: https://www.salesforce.com/blog/best-practices-for-secure-agentforce-implementation-2/

Enable Agentforce and Review Default Topics and Actions [Online]. Available: https://trailhead.salesforce.com/content/learn/projects/quick-start-create-employee-agents-in-agentforce/enable-agentforce-and-review-default-topics-and-actions

Prompt Injection [Online]. Available: https://learnprompting.org/docs/prompt_hacking/injection?srsltid=AfmBOoq66y_cUXKAf0y8wBKVrw1ZaO-8GZBQqIRsm93XVgH9QGqRUgXn

Hidden Prompts in Manuscripts Exploit AI-Assisted Peer Review [Online]. Available: https://arxiv.org/abs/2507.06185

Defending against Indirect Prompt Injection by Instruction Detection [Online]. Available: https://arxiv.org/html/2505.06311v2

Downloads

Abstract views: 0

Published

2026-06-17

How to Cite

[1]
R. Slobodian and I. Bogach, “An adversarial testing framework for AI-driven task routing systems”, Опт-ел. інф-енерг. техн., vol. 51, no. 1, pp. 374–381, Jun. 2026.

Issue

Section

Alternative Scientific Ideas and Hypotheses

Metrics

Downloads

Download data is not yet available.