AI in Compliance: From Hype to Practical Application

EQS Group, in collaboration with the German association BCM, has released a new benchmark assessing six leading AI models across 120 compliance scenarios. The "EQS Benchmark Report: AI Performance in Compliance & Ethics" reveals that AI models excel in structured tasks like classification and decision-making, achieving accuracy rates above 95%. However, performance dips notably in complex, ambiguous scenarios. Google's Gemini 2.5 Pro and OpenAI's GPT-5 lead the field, illustrating rapid advancements in AI capabilities.

Gemini 2.5 Pro scored highest overall at 86.7%, with GPT-5 closely following at 86.5%. The models performed best in structured tasks, but struggled with open-ended ones. Consistency was high, with a low hallucination rate of 0.71%, emphasizing the models' reliability when tasks are clearly defined. The benchmark highlights the significance of specific prompts and model selection in determining outcomes.

The report underlines AI's potential to enhance compliance tasks while stressing the need for human oversight, especially in sensitive areas. It encourages compliance professionals to integrate AI into their workflows, ensuring ethical and legal boundaries are respected.