Executive Summary
Policy Goal: To protect the integrity and credibility of professional accountancy
examinations in Ghana by establishing AI-resilient assessment systems that ensure examination results reflct genuine candidate competence. The Challenge: Large Language Models (ChatGPT, Gemini, Claude) now pass professional accountancy examinations at human-expert levels, rendering conventional security measures obsolete. Without urgent intervention, examina-tion results may no longer reflct authentic candidate competence, undermining professional standards and public trust in chartered accountants.
Key Finding: Experimental study of March 2024 ICAG examinations tested four AI systems against 5,698 human candidates across all three professional levels. Advanced AI models signifiantly outperformed top human candidates (p=0.019), with domain training further enhancing performance (p=0.039). All AI systems passed Level 1 courses: Level 2 and 3 performances varied by subject complexity, with highest scores in strategy-oriented papers. Advanced models substantially outperformed basic versions (p=0.029), enabling sophisticated po-tential examination fraud.
Strategic Recommendation:
• Technology Deployment: Advanced proctoring systems with AI detection capabilities across all examination levels; secure device-free examination environments.
• Regulatory Framework: Updated regulations explicitly prohibiting AI assistance with enforceable graduated sanctions; comprehensive awareness campaigns.
• Assessment Innovation: Oral examinations for Level 3 strategic papers to assess real-time professional judgment.
• Enforcement Mec-hanisms: Dedicated oversight systems to investigate suspected AI-assisted misconduct and ensure consistent compliance.
