Fighting AI with AI

Key Takeaways Copied to clipboard!

• Vocal deepfakes are highly effective scams, with millions of Americans having already lost money to AI voice fraud, often involving requests for gift cards or urgent money transfers.  Copied to clipboard!
• Businesses like Reality Defender are using AI to fight AI by detecting subtle harmonic structures in cloned voices that the human ear cannot perceive, helping banks secure phone systems.  Copied to clipboard!
• Experts argue that relying solely on voice biometrics as a password is a major vulnerability, advocating instead for multi-factor authentication layers and mandatory disclosure of AI-generated content online.  Copied to clipboard!

Segments Expand All Collapse All

Testing AI Voice Cloning

Copied to clipboard!

(00:00:11)

• Key Takeaway: A test call using a deepfake voice was conducted on a colleague who initially suspected the voice was AI.
• Summary: The hosts conducted an experiment where one speaker used a cloned voice to ask a colleague, Angel Carreras, to buy gift cards. Carreras initially identified the voice as AI, demonstrating some awareness of the technology. The test concluded with the deepfake speaker admitting the deception, highlighting the difficulty in immediate detection.

Scope of AI Voice Fraud

Copied to clipboard!

(00:01:26)

• Key Takeaway: Millions of Americans have lost money to scam calls utilizing AI voices, with individual losses often reaching thousands of dollars.
• Summary: The prevalence of successful audio deepfake scams is significant, affecting millions of Americans. These fraudulent calls often create urgent scenarios designed to trick victims into compliance. The financial impact of these scams can be substantial, amounting to thousands of dollars per incident.

Banks’ Defense Against Fraud

Copied to clipboard!

(00:03:17)

• Key Takeaway: Banks are major targets for AI voice fraud, prompting CIOs to constantly check for vulnerabilities in phone line systems.
• Summary: Financial institutions are primary targets because they hold the money fraudsters seek to access. Criminals follow digital banking trends, attempting to bypass voice verification systems used for authentication. PNC Bank’s Mark Kwapizewski emphasizes that criminals constantly probe for any security weaknesses in the bank’s defenses.

Reality Defender’s Detection Method

Copied to clipboard!

(00:04:15)

• Key Takeaway: Reality Defender detects AI voices by analyzing specific harmonic structures that are present in synthetic audio but absent in human speech.
• Summary: Ben Coleman co-founded Reality Defender after observing early stages of voice fraud while at Goldman Sachs. The company performs ‘inference’ to probabilistically indicate AI usage in audio content. Their software identifies unique harmonic structures characteristic of AI generation, even if the human ear cannot perceive them.

Layered Security vs. Voice ID

Copied to clipboard!

(00:05:52)

• Key Takeaway: While many top banks use AI detection software, experts advise removing voice authentication entirely, favoring multi-factor authentication layers.
• Summary: Ben Coleman believes banks should stop treating voice as a password, despite the use of detection services like Reality Defender. PNC Bank acknowledges the risk of relying on a single dimension like voice authentication. They prioritize layered security, incorporating location, device details, and text verification codes alongside voice checks.

Customer Vulnerability and Spoofing

Copied to clipboard!

(00:07:16)

• Key Takeaway: The greatest vulnerability in voice fraud is shifting from bank phone lines to the customers themselves, often through urgent spoofed calls.
• Summary: Criminals exploit the weakest defenses, which are now often the customers rather than the bank’s infrastructure. PNC works with telecom carriers to block spoofed numbers, preventing fraudulent calls from reaching customers. Scammers create urgency, falsely claiming accounts are compromised and instructing victims to move money to gold or cryptocurrency, which is always a scam.

Need for Universal AI Labeling

Copied to clipboard!

(00:08:44)

• Key Takeaway: Advocates are pushing for regulation that would mandate informing users whether any online content, including calls and memos, is AI-generated.
• Summary: Ben Coleman advocates for vetting all online content—text, voice, and video—for AI generation, citing issues like celebrity scams. He testified before Congress to advance regulations requiring disclosure of AI origin for content like Zoom calls or WhatsApp memos. This regulatory push aims to keep pace with rapidly advancing technology.