This UNESCO report published on 2024 highlights the rapid adoption of AI across industries and its dual nature—which may offer benefits while posing significant societal risks. The report underscores the persistence of biases in leading language models like GPT-2, ChatGPT, and Llama 2. These systems often replicate and amplify societal stereotypes, associating gendered identities with traditional or negative roles, and demonstrating biases across intersections of gender, culture, and occupation. The findings emphasize the urgent need for robust ethical frameworks to minimize harm and promote fairness throughout the AI lifecycle.

The report calls for collaborative action among governments, policymakers, and technology developers. Recommendations include adopting human rights-based guidelines, ensuring inclusive and transparent data practices, and integrating diverse perspectives in AI design and application. By embedding fairness at every stage—from training datasets to user-facing outputs—and fostering public awareness, stakeholders can work toward creating equitable and ethical AI systems that minimize harm and promote societal well-being.