In a Foreign Policy analysis, Joshua Keating argues that the more pressing near-term danger from AI is not superintelligence but “jagged intelligence,” in which models appear highly capable yet still make basic, consequential errors. He cites AI-linked mistakes in military and intelligence settings, including a hallucinated intelligence report that nearly led to a U.S. boarding of a Chinese ship. Keating warns that growing trust in these systems, combined with automation bias, makes hallucinations and poor system design more likely to cause real-world harm than speculative superintelligent takeover scenarios.
