ORCID

Abstract

Large Language Models (LLMs) may infer that they are being evaluated and strategically modify their responses to appear more aligned than they actually are; a phenomenon described as alignment faking or evaluation awareness. This presents a fundamental challenge for bias evaluation in AI systems, particularly in multi-agent workflows, where sequential decision-making steps occur with minimal human supervision. When LLMs infer that they are being evaluated, they may suppress or diminish discriminatory behaviours that could otherwise manifest in production AI Agent workflow environments, making traditional bias assessment techniques less effective.This work presents a contamination prevention architecture developed in response to observed assessment awareness during initial experimentation. Direct assessments via Model Context Protocol (MCP) revealed that LLMs were inferring bias evaluation intent from task framing, requiring the implementation of a dual presentation system that separates evaluation context from agent workflow activities.The architecture additionally addresses data contamination risks, by isolating evaluation scenarios from public exposure, preventing the incorporation of assessment data into future training datasets that would compromise assessment integrity, which is a critical consideration given documented contamination in existing AI benchmarking frameworks.

Keywords

Agentic Artificial Intelligence, Alignment Faking, Bias Assessment, Ethical AI, Human-Centred AI

Publication Date

2026-02-16

Event

3rd International Conference on Human-Centred AI - Education and Practice, HCAI-ep 2026

Publication Title

HCAI-ep 2026 - Proceedings of the 2026 Conference on Human Centered Artificial Intelligence - Education and Practice

Publisher

Association for Computing Machinery (ACM)

ISBN

9798400721533

First Page

121

Last Page

121

Deposit Date

2026-05-13

Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

Download


Share

COinS