The Illusion of Objectivity: Auditing Political Bias in Commercial AI [Level 2, 3 Constitution Ave (Nara Centre)]

​Please note: CBRIN (1 Moore Street) is currently closed and this event will be held at Collaboration Space on L2 of Nara Centre at 3 Constitution Ave. Access will be provided from the lobby entrance on the plaza side.

​When a professional uses an enterprise AI tool to synthesise research or summarise a complex policy issue, the tool feels like a fast, neutral assistant. However, commercial Large Language Models (LLMs) are far from politically neutral. In fact, if you ask major public LLMs who they would vote for in an Australian election, they will actively reveal distinct political and policy preferences.

​For organisations operating in environments where neutrality, objectivity and evidence-based decision-making are expected, relying on off-the-shelf generative tools presents a high-stakes risk. How do we ensure these automated political preferences do not quietly shape advice going to Cabinet?

​In this strategic, high-level workshop, we will step away from the technical code and look closely at the architecture of AI risk. Rather than learning how to build these tools, we will learn how to audit them. We will expose the hidden political biases in commercial LLMs and explore what more rigorous approaches look like in a public sector context.

​In this workshop, participants will:

• ​Deconstruct the Political Slant: Unpack how major commercial models express specific voting and policy preferences when tested against Australian political contexts.

• ​Run a Live Policy Stress Test: Participate in a high-level interactive exercise to identify hidden political assumptions and sycophancy in everyday AI prompts.

• ​Explore High-Dimensional Rigor: Walk through a strategic case study to see how deterministic data frameworks can map massive public sentiment with zero hallucinations and complete mathematical neutrality.

• ​Define AI Governance for High-Stakes Projects: Discuss how senior leaders and policy teams can move beyond ad-hoc experimentation and establish safe, auditable data sandboxes for complex data environments.

​This workshop is designed for policymakers, analysts, strategic leaders, researchers and reformers who want to look under the hood of generative AI and ensure important decisions remain robust, transparent and defensible.

​Facilitator

​Professor Paul X. McCarthy is a computational social science researcher and CEO of League of Scholars, specialising in mapping the technology and innovation landscape. A founding partner of the Oxford Science of Startups lab, Paul delivers trusted data science and analytics projects for the UN and OECD, state governments, and federal departments including Industry, Foreign Affairs and Trade, and Education. He co-authors The Australian’s annual RESEARCH magazine, which serves as the nation's definitive benchmark for research capability, and holds an Adjunct Professor role at UNSW.