AI Voice Fraud Outruns Detection Defenses

Relying on individuals to spot AI deepfakes is a systemic policy failure; the only way to stop voice fraud is to make banks, telcos, and AI platforms financially liable for the exploits they enable.

• –Detection-based security is dead: When the world's leading deepfake expert can no longer distinguish real voices from AI clones, any defense assuming a human or a detector can verify authenticity is obsolete.
• –Platform guardrails are mostly theater: Most consumer voice-cloning services operate on a self-attestation basis, lacking technical verification of consent and offering trace-back tools only after a crime has occurred.
• –Financial liability is the ultimate lever: As shown by the UK's APP fraud reimbursement rules, forcing institutions to share the cost of scams is the most effective way to drive the adoption of protective transaction friction and stop fraud.