How Financial Institutions Can Combat AI-Deepfake Fraud

Do you ever wonder how security teams stay ahead of constantly evolving hacker tactics? Or how different organizations have the same plan to mitigate adversary behaviors? Security teams leverage MITRE ATT&CK to gain insight into attacker behavior. In 1958, MITRE was established to help advance technology and bridge the academic community to industry. It wasn’t until 2013, where Mitre launched MITRE ATT&CK to better protect against cyberthreats. What started as an experiment of real-world observation soon became a framework that many security teams at leading organizations leverage to safeguard their assets. Today MITRE ATT&CK helps security teams in all sectors, and empowers them with objective insights to detect known adversary behaviors. This framework allows all stakeholders involved, from security teams to vendors, to clearly state the threat assessment and the plan to help solve it. 

MITRE ATT&CK's framework has become the go-to resource for understanding and responding to adversary tactics, techniques, and procedures (TTPs). It offers a structured approach for mapping out attack vectors and providing consistent terminology that organizations can use to communicate threats. This framework supports proactive defense measures, enabling companies to identify gaps in their cybersecurity posture, prioritize responses to emerging threats, and continuously improve their defenses based on real-world data.

Step-by-Step Breakdown of AI Voice Fraud Techniques 

Understanding the specific steps that attackers take can help organizations better prepare and defend against these sophisticated schemes. This section provides a detailed example of the stages involved in an AI voice fraud attack against a bank, from initial reconnaissance to the final exfiltration of sensitive data. 

Reconnaissance:

The attacker begins by gathering publicly available information about the target, such as the names of key people, company structure, and customer service protocols. Social media, LinkedIn profiles, and leaked data can help them understand who to impersonate and what information might be valuable.

Resource Development:

To support their attack, the attacker uses generative AI to create high-quality voice clones of key individuals in the organization. For example, they might create a voice clone of a senior executive by using accessible voice data from public speeches, webinars, or podcasts. Additionally, they might acquire tools for spoofing caller IDs to appear as if calls are coming from within the company.

Initial Access:

The attacker then uses the cloned voice to call the company’s customer service or a lower-level employee to initiate the attack. They might impersonate a senior executive, claiming to be locked out of a critical account or urgently needing assistance with an issue. The goal here is to convince the supporting team member to help them gain access to internal systems or sensitive accounts.

Execution:

After gaining initial access, the attacker may proceed to call other departments to inquire about other information. For instance, they could impersonate the CFO to inquire about specific financial accounts, requesting verification of recent transactions or even asking for wire transfers.

Persistence: 

Continuing to attack specific high-profile users with vishing will allow for a hacker to have stronger access to a companies accounts. 

Privilege Escalation: 

Using the voice they are impersonating, the attacker escalates their requests, asking for more detailed information or access privileges. For instance, they might ask IT to reset passwords or ask someone on the finance team to move money to specific accounts, moving from general information to more sensitive and critical system access.

Defense Evasion:

To avoid detection, the attacker can use different spoofed phone numbers and create subtle changes in voice tones to simulate natural phone quality variations. If questioned, they may provide convincing excuses, such as being in a noisy location, to cover any discrepancies.

Credential Access:

The attacker aims to collect login credentials, social security numbers, financial account details, and other sensitive information, which they can use to further exploit the organization.

Discovery:

The attacker uses the access they’ve gained to explore the organization’s systems and identify valuable resources, such as sensitive data and privileged accounts.

Lateral Movement:

With the information collected, the attacker moves within the organization’s systems, accessing additional departments or servers. For example, they may use the compromised credentials to log into other financial systems or corporate databases.

Collection:

The attacker gathers data relevant to their goal, such as financial reports, customer data, or intellectual property. In the case of ransomware, this data could be encrypted; if the goal is data theft, it might be exfiltrated.

Exfiltration:

Sensitive data is transferred out of the company’s network, potentially over encrypted channels to avoid detection by the organization’s security systems.

Impact:

The attacker may disrupt business operations by stealing money, modifying critical financial records, or even launching a ransomware attack, demanding payment to restore access to the compromised data. Even then, the stolen information could be sold on the dark web or used for further financial gain.

This example demonstrates how generative AI voice fraud can exploit the human factor in security, bypassing traditional technical defenses and causing significant harm through social engineering tactics.

Real-World Examples of AI Voice Fraud

Real-world incidents of AI voice fraud highlight the alarming potential to facilitate cyber attacks. One notable case occurred in 2019, when criminals used AI-generated voice technology to impersonate a German energy company's CEO. By mimicking the CEO's voice and tone, they successfully convinced a senior executive to transfer $243,000 to a supplier, exploiting the executive's trust in the apparent authenticity of the call.

Another incident occurred in Hong Kong, where a bank suffered a $35 million loss due to a deepfake voice scam. Fraudsters used AI-generated voice technology to impersonate a director and authorize the transfer of the funds. The incident showcased how sophisticated AI techniques can bypass traditional verification methods, exploiting the human element in cybersecurity.

In another instance, scammers used AI-generated voices to stage a fake kidnapping, contacting a target's family with a fabricated audio of the victim's distressed voice. The attackers demanded a ransom, leveraging the convincing nature of the voice clone to pressure the family into compliance.

These examples illustrate the evolving threat posed by AI-enabled voice fraud, demonstrating how attackers can bypass conventional security measures and manipulate human trust. Organizations must strengthen defenses against such social engineering tactics by employing advanced detection techniques, multi-factor authentication, and continuous employee awareness training.

By understanding these cases, companies can better prepare to detect and mitigate the risks associated with AI-driven cyber threats.

The Future of Voice Fraud and AI in Cybersecurity

As AI technology advances, the capabilities of generative AI will continue to grow, enabling attackers to produce even more realistic and convincing voice clones. Future iterations of AI-driven voice fraud may involve varying voice characteristics, allowing attackers to adjust their responses dynamically during a live conversation, making it harder to detect. Additionally, AI may be used to analyze voice data for emotional cues, enabling attackers to manipulate their targets more effectively by exploiting psychological triggers.

These advancements will pose new challenges for cybersecurity in the future. Traditional defenses may struggle to keep up with the evolving techniques, and the risk of deepfake voice fraud becoming mainstream could increase, especially in industries like finance, healthcare, and other industries where trust in voice communication is crucial.

To counter these threats, companies are developing new measures and standards for AI usage and voice-based authentication. Potential measures might include mandating companies to implement multi-factor authentication for voice-based processes, requiring the use of voice biometrics to validate identities, and setting guidelines for the ethical use of AI technologies. These regulations could help establish industry-wide standards and ensure companies remain accountable for safeguarding sensitive information.

Have you Herd?

AI-driven voice fraud is a sophisticated and evolving threat, but organizations can stay one step ahead with the right defenses. Herd Security’s advanced voice detection is designed to combat these types of social engineering attacks by identifying suspicious behavior patterns and voice anomalies. Leveraging insights from frameworks like MITRE ATT&CK, Herd Security helps organizations detect potential voice fraud at various stages—whether during the initial access phase or when attackers attempt to escalate privileges. Our solutions empower security teams to proactively mitigate the risks, continuously adapt to new threats, and strengthen the human element of cybersecurity.

Don’t let AI-powered attacks catch you off guard. Stay vigilant and protect your organization with Herd Security. Be a member of our pilot program today.