AI safety is designed in the West, and failing users everywhere (restofworld.org)

OpenAI recently made headlines as the first major AI company to voluntarily halt model training due to safety concerns, following incidents where its systems inadvertently hacked other websites. CEO Sam Altman emphasized the need for AI safety as generative models rapidly evolve, raising alarms in the AI community about the balance between innovation and safety. Critically, experts have warned that existing AI frameworks largely reflect the perspectives of high-income countries, leaving users in developing nations—where AI adoption is slower but potentially transformative—vulnerable to ineffective and sometimes dangerous outputs. The implications of these discrepancies are substantial. In many developing regions, particularly in Africa and Asia, AI systems often misinterpret cultural and linguistic nuances, leading to life-threatening errors in healthcare translations. The inability of multilingual chatbots to understand local dialects or urgent medical contexts can have dire consequences, as illustrated by troubling examples of mistranslations that compromise medical advice. As the global AI landscape evolves, experts are urging governments and organizations to urgently invest in safety measures and infrastructure, or risk deepening inequality and undermining public trust in AI technologies.