Wall Street vishing attacks expose voice as a weak link in identity security

A wave of voice-phishing attacks targeting some of Wall Street’s largest hedge funds and private equity firms is exposing a weakness that stronger passwords, multifactor authentication and other identity safeguards cannot fully address. Employees can still be tricked into believing they are speaking with someone they trust.

Point72 told investors that no customer information had been stolen, while Two Sigma said it had no indication that its data or systems were affected. Citadel declined to comment on whether its systems had been breached.

The attacks appear to have been part of a considerably larger campaign. Google Threat Intelligence Group said August 6 that a threat actor it tracks as UNC6671 had shifted its targeting toward financial services, private equity and professional services while operating through several extortion brands, including Redact, Pink, Helix and Falcon.

Reuters subsequently linked numerous financial firms to the campaign by analyzing company-specific malicious domains associated with infrastructure identified by Google.

Reuters, using malicious domains identified by Google and Internet intelligence services, found infrastructure apparently prepared to target Blackstone, Bridgewater Associates, Apollo Global Management, Bain Capital, KKR, TPG, CME Group, Clearlake Capital and Moody’s.

The attackers had created digital traps for more than 200 companies in roughly five weeks.

The campaign illustrates how attacks on identity systems increasingly begin outside the identity system itself.

Google said UNC6671 callers pose as corporate IT or help desk personnel, sometimes calling employees on their personal cellphones and in some cases spoofing a company’s legitimate help desk number.

The caller creates urgency around what appears to be a routine security task, such as enrolling a FIDO2 passkey or updating multifactor authentication (MFA).

The employee is then directed to a convincing but fraudulent authentication site bearing the targeted company’s name.

As the employee enters a username and password, an adversary-in-the-middle system captures the credentials and submits them to the real identity provider.

When the legitimate service requests a second authentication factor, the attacker attempts to capture that as well.

Once inside, Google said the attackers have registered their own MFA devices to maintain access and used compromised accounts to move through Microsoft 365 and Okta environments, reaching SharePoint, OneDrive and connected applications such as Zendesk and Salesforce.

Automated scripts can extract large amounts of data before the victim is approached for extortion.

That makes the campaign an identity attack as much as a conventional phishing operation. It is designed to persuade the legitimate user to help the attacker defeat controls intended to establish who is seeking access.

AI adds another dimension. Gartner characterized the recent hedge fund incidents as AI-powered voice phishing and warned that organizations need broader controls around business processes that depend on telephone communications.

Other reports have described attackers using or being capable of using AI-generated voices.

Google’s detailed description of UNC6671’s tradecraft, however, centers on impersonation, spoofed help desk numbers, fraudulent authentication sites and real-time interception of credentials and MFA tokens.

Google does not say in its reports that investigators confirmed the use of cloned or AI-generated voices in the calls.

That distinction does not diminish the problem posed by increasingly realistic synthetic speech. Cheap voice-generation systems make it possible to imitate the characteristics people have traditionally relied upon to recognize another person, while automating attacks that once required considerably more human effort.

Ben Colman, co-founder and CEO of deepfake detection company Reality Defender, said the Wall Street attacks demonstrate why voice itself can no longer function as evidence of identity.

“A familiar voice on a call is no longer proof of who is on the line,” Colman wrote in an August 10 analysis of the attacks.

His larger point is that financial institutions have spent years hardening digital authentication while continuing to treat the human voice as an implicit authentication mechanism.

Email systems added protections against spoofing while corporate logins added multifactor authentication and rules that can block suspicious access. Sensitive data transfers were increasingly encrypted and logged.

Phone calls, however, continued to rely heavily on a much older form of trust. If the voice sounded familiar, people often assumed they knew who was speaking.

Synthetic audio erodes that assumption. Even without a cloned voice, the UNC6671 campaign demonstrates how a convincing caller armed with organizational information, a spoofed telephone number and a plausible security pretext can move an employee into an attacker-controlled authentication process.

Help desk processes present another vulnerability. UNC6671 has exploited employees’ trust in internal IT by posing as help desk personnel. Separately, Colman notes that actual service desks deserve protection because they routinely deal with people who are having trouble authenticating themselves.

Google recommends phishing-resistant authentication such as FIDO2 security keys and passkeys, along with tighter session controls, restrictions on authentication from unmanaged devices and suspicious networks, and monitoring for abnormal MFA enrollment and large-scale cloud data access.

Properly implemented FIDO2 and WebAuthn authentication can also defeat the lookalike sites used in adversary-in-the-middle attacks because the authentication credential is cryptographically bound to the legitimate website.

But preventing credential theft addresses only part of the problem. Financial firms also must reconsider transactions and security decisions that can be authorized simply because someone calling appears to sound like a known executive, colleague or IT employee.

Colman argues that synthetic media detection can become another security signal inside calls and video meetings, while emphasizing that it should not itself become the identity credential.

“Detection is a layer, not authentication,” he wrote.

The Wall Street campaign points toward the need for a broader change in identity security, because it can no longer focus solely on passwords, devices, tokens and other conventional authentication controls.

Organizations now must consider whether the person making a request is genuine and whether the communications channel carrying that request can be trusted.

Article Topics

deepfake detection  |  deepfakes  |  identity security  |  Reality Defender  |  synthetic voice  |  UNC6671  |  voice biometrics