MIT and University of Queensland study: Large corporations remain silent on AI risks

A study by MIT and the University of Queensland analyzed corporate disclosures of AI risks. Among the top 100 companies, 32% mentioned no AI risks at all; estimates suggest 62% among the top 1,000. The most severe risks were least mentioned: none of the top 100 discussed power concentration or AI arms races, and only four cited dangerous AI capabilities. AI developers reported hallucinations in 15% of cases, while bias was noted in 85% (based on data from just 13 developers). The authors argue companies should explicitly list their own risks and corresponding mitigation measures.