How Brand Protection Can Address the Risk of GAN Deepfakes
Deepfakes are a concept that has taken root in popular culture. Most deepfakes are benign; the good ones go viral and can often make us laugh. But in the very near future, deepfake attacks waged against businesses will be unlikely to put a smile on anyone’s face. I’ll describe the technology behind deepfakes, known as a generative adversarial network (GAN), the risk of attacks against organizations and present a multilayered strategy for defending against them.
What is a Generative Adversarial Network (GAN)?
A generative adversarial network (GAN) is a machine learning (ML) model consisting of two neural networks:
● A generator—A convolutional neural network. This model artificially manufactures outputs that can be mistaken for real data.
● A discriminator—A deconvolutional neural network. This model identifies which of the outputs it receives were created artificially.
The generator and discriminator compete with each other to improve the accuracy of their predictions. As the feedback loop between the networks continues, the generator produces higher-quality output and the discriminator becomes better at flagging artificially created data.
This process enables the GAN to create its own training data, which is why GANs are usually unsupervised, learning through a cooperative zero-sum game framework.
What are Deepfakes?
Deepfake is an artificial intelligence (AI) program that can create believable audio, video and image hoaxes. The term deepfake combines two terms—deep learning and fake—to describe the deep learning technology creating fake content.
Deepfake content is created by two competing models. The generator creates fake multimedia content and asks the discriminator to determine between artificial and real content. Together the models for a GAN that continuously improves the deepfake content.
Establishing a GAN requires identifying the desired output and creating a training dataset for the generator. After the generator creates an acceptable output quality, you can start feeding artificial content to the discriminator.
As the generator creates more believable fake content, the discriminator better distinguishes between fake and real content. As the discriminator gets better at identifying fake content, the generator gets better at creating it. And so the loop goes on.
The Risk Deepfakes Pose for Businesses
Most businesses have measures to protect their information’s confidentiality, integrity and availability (CIA). Data breaches frequently threaten data confidentiality, and ransomware attacks increasingly target data availability. These are known threats.
Data integrity seems to be a more recent phenomenon, perhaps because it requires more advanced technology, but it can have devastating impacts on businesses. As deepfake technology improves, it enables threat actors, including competing businesses, nation-states, criminals, anonymous saboteurs and disgruntled employees, to launch attacks easily.
Deepfake-Driven Voice Phishing (Vishing)
Phishing is a form of fraud that impersonates brands or trustworthy entities to trick victims into divulging information, clicking on malicious links and performing various nefarious actions unwittingly. In late 2019, the Wall Street Journal reported that a deepfake audio technology was used to mimic a CEO’s voice and convince a company underling to transfer funds fraudulently.
Viral Content
Deepfake content can severely impact the reputation of a business or individual. They can quickly go viral when posted on social media—spreading worldwide in mere minutes. To repudiate the legitimacy of a deepfake, businesses need to expend many resources to identify, remove and refute it, in addition to wasting resources on crisis management and legal fees.
A deepfake embedded in corporate systems by an insider threat requires the business to investigate the network intrusion incident and remediate all corrupted systems and data. Though the business might eventually prove the offending content was a deepfake, the reputational damage is done by that time and the business might suffer lost revenues.
What is Digital Risk Protection?
Digital risk protection (DRP) software helps secure digital assets, using cyberthreat intelligence (CTI) monitoring to recommend actionable protections. DRP platforms use intelligent algorithms and reconnaissance methods to locate, track and analyze threats in real-time and can feed this information into automated response solutions.
A DRP solution uses indicators of compromise (IOCs) and indicators of attack (IOAs) to analyze risks and alert security teams when identifying potential attacks. It provides data handling and analysis capabilities that highlight important information to prevent teams from becoming overwhelmed by intelligence data to the point they overlook critical threats.
How DRP Can Help
Preventing and countering deepfakes may involve several technological and practical measures. Security best practices can help organizations avoid deepfake-related fraud by integrating automated checks into all relevant processes and helping employees identify potential videos.
For example, extra precautions can help secure payment processes by making it harder to fool employees and pass all security controls.
Businesses can incorporate deepfake-focused security controls into their DRP systems. The DRP technology can help reduce the manual burden of sorting through content and preventing malicious actors from inserting deepfakes into the system.
A solid security strategy to combat deepfakes should include the following elements:
● Employee training—Everyone in the organization should be aware of the risk posed by deepfakes, understand how deepfake-driven attacks work and how to recognize deepfake content.
● Secure business processes—Fundamental procedures should use a “trust but verify” approach. Maintaining a suspicious attitude towards video and voicemail content does not guarantee that an individual catches every scam, but it can help reduce the likelihood.
● Deepfake identification—This is where the DRP solution becomes especially useful. Automation makes it easier to reliably identify an attack, especially when using AI-driven detection software. Deep learning solutions can recognize signs of tampering in images or videos.
● Watermarks—Applying watermarks to critical visual data helps prevent tampering and provides evidence if someone has doctored the image.
● Blockchain—This decentralized solution lets users store data online without requiring a central server. Blockchains can avoid various security vulnerabilities that affect central data storage systems. While distributed ledgers are a practical solution for storing large volumes of data, they can also securely store electronic signatures and hashes.
Incident Response
Whatever the specific practices implemented to detect and respond, it is essential to have a well-planned security strategy to prepare teams for an attack. When an employee or the DRP system discovers a deepfake, the organization can react in the right way.
The strategy should specify the activities and responsibilities assigned to teams and individuals. It should be multi-layered and cover critical security aspects, including:
● Identity checks, including device ID, validation and analytics.
● Activity monitoring and behavioral analytics
● Document verification
The first protection layer ensures that only authorized individuals can access vulnerable resources. Additional layers provide visibility, enforcement, analysis and security controls against attackers.
Conclusion
It sounds scary, and while the threat of deepfake attacks against businesses is greater than ever, it’s not insurmountable. With a combination of staff awareness, robust DRP solutions with detection capabilities and an organizational process to respond to attacks, you will be prepared. At least, until threat actors discover the next holy grail of technological deception.
