What U.S. Threat Assessments Say About AI Risks

Summary

This opinion essay argues that public debate over an AI apocalypse often obscures the more concrete risks described by U.S. government assessments. The intelligence community’s annual threat assessment says humans must retain control, that risks from AI autonomy need mitigation before broad deployment, and that AI may significantly affect national security; it does not warn of human extinction or a machine uprising. The Department of Homeland Security lists five concerns: disinformation and election interference, cyber operations and financial crime, extremist exploitation, access to chemical and biological knowledge, and attempts to break AI safeguards or poison training data. The author argues that most are longstanding threats that AI can increase in speed or scale. Citing RAND, the essay characterizes AI as adding “entropy” to existing problems and notes that catastrophic scenarios such as engineered pathogens, geoengineering, or nuclear war face physical and operational constraints. A RAND study found no statistically significant improvement in plans for large-scale biological attacks when AI was used. The National Academies identifies practical bottlenecks, including DNA screening, laboratory skills, and difficult production steps that require human intervention. The author concludes that these risks deserve attention, but the cited assessments do not support claims that AI makes human extinction imminent.