MIT Technology Review reports AI-driven scams have evolved from LLM-generated phishing post-ChatGPT to deepfakes and automated scans, while healthcare AI for notetaking, record analysis and imaging lacks clear evidence of improved patient results.

Cybercriminals adopted generative AI after its 2022 public release, reducing attack costs and increasing volume beyond organizational defenses (Rhiannon Williams, MIT Technology Review, 2026). Chainalysis documented AI-enhanced fraud rising over 200% in cryptocurrency scams during 2024-2025, with LLMs enabling personalized campaigns at scale (Chainalysis 2025 Crypto Crime Report). IBM's 2025 Cost of a Data Breach Report cited AI as a factor in 43% faster incident response times for defenders yet faster attack execution (IBM, 2025).

Studies confirm AI tools achieve high diagnostic accuracy on X-rays and records, yet randomized trials show no translation to lower mortality or better recovery (Jessica Hamzelou, MIT Technology Review, 2026; JAMA Internal Medicine, 2025). A New England Journal of Medicine analysis of 32 deployed systems found only 9 included outcome evaluations beyond technical metrics (NEJM, 2024).

Mainstream reporting treats AI cyber risks and healthcare applications as isolated; primary sources including NIST AI RMF 1.0 updates link both to insufficient pre-deployment testing on real-world impacts versus capabilities (NIST, 2023; MIT Technology Review, 2026).