Pindrop claims up to 99% accuracy detecting synthetic voice fraud plaguing call centers

Pindrop has released its 2024 Voice Intelligence and Security Report, and its conclusions will come as no surprise to anyone paying attention in the biometrics and digital identity sectors. Data breaches are on the rise, fraud is evolving as audio deepfake tools get better and easier for fraudsters to access, generative AI is penetrating into the mainstream, and synthetic identities are everywhere.  There is good news too, however, in success rates Pindrop has found catching deepfakes with biometric liveness detection.

“Fraudsters have discovered that deepfakes, with the aid of mature generative artificial intelligence tools such as ChatGPT, can create more targeted and individual-specific attacks that can spread misinformation, extract funds, or harm an individual’s reputation,” says the report. “The new GenAI revolution, while promising to enhance commerce and human productivity, also threatens to sow fundamental distrust in companies, institutions, and online media.”

Armed with sophisticated speech models and leaked identity credentials obtained from data hacks or breaches, often available for sale on the darkweb, fraudsters are increasingly employing tactics such as targeted smishing, social engineering, robotic dialing and impersonation tactics. Emphasizing in a press release that “generative AI fundamentally breaks trust in commerce, media, and communication” Pindrop CEO Vijay Balasubramaniyan wants to fight fire with fire. “We need good AI to beat bad AI,” he says.

Phones still largest channel as fraudsters blitz contact centers

Major targets for fraudsters include executives, social media platforms – and, crucially for Pindrop, contact centers. With many consumers still relying on phones to complete transactions, call centers are vulnerable to voice spoofs, audio deepfakes and other fraud attacks. “Contact center fraud started spiking in 2022 after a brief slowdown in 2021,” says the report. “The fraud rate has increased from one in every 1200 calls in 2021 being fraudulent to one in every 756 by the end of 2023.”

Additionally, fraudsters use social engineering tactics to deceive customer service representatives, persuading them to remove restrictions or reset online banking credentials. And AI bots are increasing in complexity, able to conduct simple conversations that can lead to more serious account activities.

Banks, credit unions and insurance firms are at risk, but the impact is not isolated to financial services; social media is a target, and e-commerce has enabled rampant retail fraud. Pindrop says up to 14 percent of all e-commerce returns are fraudulent. “We observed a 4x rise in the retail fraud rate through 2023 in our customer call centers as more instances of concessions abuse and fraudulent returns have come to light, particularly in the contact center.” Moreover, retail fraud is projected to more than double in 2024.

“Call centers are the backbone of customer interactions,” says the report. “Retailers will need to take steps to keep them secure by adopting advanced voice authentication and fraud detection policies.” A large section on multifactor authentication (MFA) summarizes the cost and security benefits of implementing MFA: “In 2023, the implementation of MFA solutions yielded remarkable results, as evidenced by a group of 33 Pindrop customers collectively saving approximately $25 million in handle time cost reductions.”

Liveness backs up Pindrop’s 99% detection rate

Another major factor in fighting fraud is liveness detection. With numbers showing higher accuracy for liveness detection than human ears and voice biometrics, it is among the best ways to authenticate real human interactions – and is strongest when combined with multiple factors, posting a 99.2 percent deepfake detection accuracy rate.

Humans are only able to detect voice deepfakes a little over half of the time. Pindrop says its voice biometrics alone are 88 percent effective at identifying deepfakes, and its liveness detection technology on its own is effective 98.3 percent of the time, according to the report.

“The first order of defense now needs to be about verifying a live and real human presence, whether through voice, facial, or other identifiers,” says the report. “A deepfake, by itself, may not be an indicator of fraud. Still, there should be enough pause to detect a non-live interaction and formulate a fast and consistent response toward security and customer service.”

Pindrop Chief Product Officer Rahul Sood notes that “Pindrop’s deepfake detection technology has demonstrated a 99 percent detection rate with a false positive rate of less than 1 percent. So our customers should be confident that very few deepfake frauds, if any, will get through in the first place.”

ISMS report highlights third-party and vendor risks

Another report, this one from ISMS.online, surveys the state of information security, and finds that deepfakes are named as the second most frequent information security concern for businesses in the UK.

A summary of the report from Microsoft says “scammers are most likely to deploy deepfake content in business email compromise (BEC)-style attacks with AI-powered voice audio and video cloning used to deceive recipients into complying with the threat actors’ request, usually large volume fund transfers.” Evidence bears this out, with the revelation last week that the target of the Hong Kong deepfake attack that led to a loss of $25 million was UK sustainable development consortium Arup.

Thirty-eight percent of respondents say managing third-party and vendor risks are their businesses’ most significant challenge. “This is a direct consequence of expanding digital ecosystems and increasingly interconnected business operations. Effective management of these risks is crucial for minimizing the attack surface and ensuring the integrity

and reliability of business operations.” Regulatory compliance was also highlighted as a challenge.

Article Topics

biometric liveness detection  |  biometrics  |  deepfake detection  |  Pindrop  |  synthetic voice