AI risk: crisis and fraud

Sen. Elizabeth Warren said on April 22 that an artificial intelligence bust could spill past Silicon Valley and hit the broader U.S. financial system, echoing the chain reactions of 2008. (theverge.com) Warren delivered the warning at Vanderbilt Policy Accelerator’s “The Looming AI Crisis” event, where she said a debt-driven artificial intelligence bubble had become “a structural risk” to economic and financial stability. (senatebankingdemocrats.substack.com) In prepared remarks released the same day, Warren said artificial intelligence companies were seeking taxpayer support and guarantees if conditions deteriorate, and she called for “simple structural reforms” before any crash. (publicnow.com) A separate warning is coming from fraud specialists, who say the damage is already showing up in payments rather than in a market crash. Fortune’s cybersecurity section on April 22 argued that artificial intelligence-enabled fraud is slipping through financial systems “one authorized transaction at a time.” (fortune.com) That phrase points to authorized push payment fraud, a scam in which a customer is manipulated into sending money themselves, often after a fake bank alert, tech-support call, or investment pitch. Deloitte said in October 2025 that these transactions are hard to reverse because the victim appears to have approved them. (deloitte.com) Deloitte estimated U.S. losses from authorized push payment fraud could rise to $14.9 billion by 2028 from $8.3 billion in 2024, and said losses could reach $18.2 billion in a more aggressive scenario where artificial intelligence outpaces institutional defenses. (deloitte.com) Industry surveys show the pressure is broader than one scam type. Nasdaq Verafin said in its March 2026 global financial crime report that 90% of financial professionals it surveyed had seen an increase in artificial intelligence-driven attacks over the prior two years. (verafin.com) The Association for Financial Professionals said on April 14 that 76% of U.S. organizations experienced attempted or actual payments fraud in 2025, while only 17% were using artificial intelligence for fraud mitigation. The same survey found 74% were hit by business email compromise. (financialprofessionals.org) Banks and regulators have spent much of April focused on whether advanced models could expose software flaws and trigger a systemic cyber event. The fraud warnings describe a slower failure mode: losses accumulating through customer-approved transfers that standard controls often treat as legitimate. (fortune.com) (deloitte.com) Warren’s speech and the fraud data land on the same point from opposite directions: firms rolling out artificial intelligence face pressure not only to build faster, but to police leverage, payments, and basic controls before the next failure scales. (senatebankingdemocrats.substack.com) (financialprofessionals.org)