di_blog_post

Deep Instinct prevents more advanced threats than any EPP or EDR in the world.

Deep Instinct Blog: Breaking News and Updates

Prevention Platform

Prevention for Applications

Prevention for Endpoints

Products

Prevent Ransomware

Prevent Zero-Day Attacks

Improve EPP Efficacy

Extend & Enhance EDR

Stop Fileless Attacks

Solutions

Deep Learning in Cybersecurity

Our Customers

+ MS Defender

VS Traditional Antivirus

Why Deep Instinct

About Deep Instinct

Newsroom

Careers

Contact Us

Company

Asset Library

Blog

Videos

Training

Resources

Request Demo

Customer Portal

Integrations and Compliance

Quick links

Events & Webinars

Leadership Team

Board of Directors

Partners

Login

content_block

Perspective & Findings

ChatGPT and Malware: Making Your Malicious Wishes Come True

chatgpt-malicious-wishes-blog.png

Listing preview

Perspective & Findings

ChatGPT and Malware: Making Your Malicious Wishes Come True

image_media_card

Malicious JARs and Polyglot files: “Who do you think you JAR?”

New MuddyWater Threat: Old Kitten; New Tricks

Polonium APT Group: Uncovering New Elements

Emotet’s Vacation is Over: No Rest for the Wicked

LSASS Memory Dumps are Stealthier than Ever Before

LSASS memory dumps

Malicious Office files: 20+ Years of Microsoft Office Exploits

MIT Explores How Deep Learning is Transforming Cybersecurity

Trending at Deep Instinct

carousel_cards

listing_preview

This article was originally written for the Amazon Partner Network Blog Post

&nbsp;

At Deep Instinct, we apply end-to-end deep learning to cyber

APN-blog.jpg

This article was originally written for the<a href="https://aws.amazon.com/blogs/apn/how-deep-neural-networks-built-on-aws-can-help-predict-and-prevent-security-threats/" target="_blank" rel="noopener noreferrer"> Amazon Partner Network Blog Post</a>
&nbsp;
At Deep Instinct, we apply end-to-end <a href="https://www.deepinstinct.com/">deep learning to cybersecurity</a>.
Deep learning is inspired by the human brain. Once a brain learns to identify an object, its identification becomes second nature. Similarly, as Deep Instinct’s artificial neural network learns to detect more and more types of cyber threats, its prediction capabilities become instinctive. As a result, malware both known and new can be predicted and prevented in zero-time.
<a href="http://www.deepinstinct.com/">Deep Instinct</a> is an <a href="https://aws.amazon.com/partners/">AWS Partner Network</a> (APN) Select Technology Partner whose <a href="https://www.deepinstinct.com/">predictive threat prevention</a> platform can be applied against known or unknown threats, whether it be a file or fileless attack. It works completely on raw data to classify between malicious or benign files and can be applied to any device and all major operating systems.
As the critical infrastructure behind Deep Instinct’s deep learning network architecture, Amazon Web Services (AWS) supports the training of the deep learning neural network and drives real-time threat intelligence, file reputation, and event analysis.
This post addresses how Deep Instinct uses AWS services to process millions of queries per day and to support the huge processing power involved. Dealing with such a large number of data samples in each training cycle can be challenging, but AWS allows us to operate an architecture that supports millions of endpoints with strict KPIs and SLAs for performance and availability.
Creating Datasets for Training the Deep Neural Networks
The accuracy of Deep Instinct’s inference model necessitates the ability to train on as many data samples as possible to ensure enhanced reliability and resilience. It’s important to note the reliability of the inference models becomes incrementally better with the number of iterations that have been conducted to improve the model.
The dataset is created from multiple sources and stored in <a href="https://aws.amazon.com/s3/">Amazon Simple Storage Service</a> (Amazon S3). From each of these files, Deep Instinct uses the raw data to support the training cycles so the <a href="https://www.deepinstinct.com/">neural network</a> can perform non-linear correlations.
This creates a significant computing challenge, though, as it necessitates computing power that can’t be facilitated in-house using generally available CPUs and memory constraints in pre-training stages. The entire training phase, therefore, is conducted in-house by Deep Instinct using GPUs.
For this purpose, <a href="https://aws.amazon.com/ec2/">Amazon Elastic Compute Cloud</a> (Amazon EC2) is utilized to create Spark clusters. Spark is a distributed computing framework for big data processing and analysis.
Amazon EC2, meanwhile, offers the ideal infrastructure to host these clusters, and often serves as an expandable, low-configuration service that’s an easier alternative to running in-house cluster computing. The Amazon EC2 machines are used as <a href="https://aws.amazon.com/ec2/spot/">Amazon EC2 Spot instances</a>, which enables Deep Instinct to keep costs low while providing business continuity.
Using Spark allows for linear scalability, which is conducive to growing the number of training samples. Even if the datasets grow in orders of magnitude, the training process is minimally affected and can keep abreast without changing the inherent code to accommodate the increased load.
Supporting Deep Instinct’s Threat Intelligence
According to the <a href="https://www.all-about-security.de/fileadmin/micropages/Fachartikel_28/2019_Cost_of_a_Data_Breach_Report_final.pdf">Ponemon Institute,</a> enterprises waste an average of 425 hours a week responding to and investigating false positives, costing them an average of $1.37 million per year.
Furthermore, a study by <a href="https://www.helpnetsecurity.com/2019/05/03/businesses-security-breaches-increase/">Vanson Bourne</a> which polled 600 IT decision-makers (300 from IT operations and 300 from IT security), indicates that 55 percent of CISOs consider their solution’s verdicts to be unreliable or wrong. These poor results attest to the industry-wide need for solutions that provide high detection rates and the lowest possible false positives.
False positives are widely considered to be an unavoidable challenge for many cybersecurity professionals, so the ability to effectively avoid these enables security staff to focus efforts on more productive activities.
To meet this need for customers, Deep Instinct prioritized the investment in our file reputation services. A separate repository was required for storing events and file scanning data from production environments to further reduce the rate of false positives.
This second database was hosted in just one location so that data would be searchable and easily correlated with existing data from our AWS-based <a href="https://www.deepinstinct.com/deep-instinct-multi-layer-protection/">D-Cloud reputation layer</a>. This cloud-based environment stores file records, threat intelligence, prevalence, event information, and rich metadata on billions of files from dozens of different sources.
That said, the scalability of Deep Instinct’s solution was also important so it could facilitate the constant inquiries coming in each and every day. A normal week could see thousands of inquiries from Deep Instinct’s various agents around the globe. A centralized server was necessary, as responses have to be fast.
There are billions of files that comprise the full spectrum of file types and platforms, be it PE, PDF, OLE, APK, Fonts, RTF TIFF, SWF, or Mach-O. This enormous variation means any previously known malware has been labeled with its respective verdict.
D-Cloud provides a comprehensive file reputation service that further reduces false positives and improves Deep Instinct’s threat intelligence infrastructure. This infrastructure supports endpoints around the globe so that no matter where an agent may be located it can inquire D-Cloud and receive the respective verdict in less than 200 milliseconds.
Stages of Verifying Data Inquiries
Deep Instinct’s threat intelligence database on AWS receives millions of new samples each week from a global pool of sources, making it a trusted source of information that can be used to verify all types of data inquiries.
The agent (D-Client in Deep Instinct’s case) can query the <a href="https://aws.amazon.com/dynamodb/">Amazon DynamoDB</a> database for details on a specific sample via <a href="https://aws.amazon.com/lambda/">AWS Lambda</a>, which fetches the details and passes it on to the agent.
The serverless Lambda function is a small piece of software that has all of the logic inside, including protocols and methods to fetch the data from the database. AWS Lambda relies on <a href="https://aws.amazon.com/cloudfront/">Amazon CloudFront</a>, which is a fast, highly secure, and programmable content delivery network (CDN).
CloudFront can update or upgrade the D-Client (end device) in the fastest way available, as AWS installs CloudFront servers in locations in the world so users can access a server that’s localized to their geographic location.
<a href="https://aws.amazon.com/lambda/edge/">Lambda@Edge</a> acts as a protocol between the two components of the software. Prior to the request, the protocol needs to be defined between the agent and DynamoDB on how it should be queried and how the response is to be relayed back.
The Lambda is created in the cloud, and the agent team is provided with all the details on how to query data on files.
Customer Use Case: Designing for Millions of Devices Worldwide
Deep Instinct <a href="https://www.deepinstinct.com/2019/05/22/hp-collaborates-with-deep-instinct-to-roll-out-ai-powered-malware-protection-for-next-generation-hp-elitebook-and-zbook-pcs/">collaborated with HP</a> to roll out an artificial intelligence-powered malware protection application called <a href="https://www.deepinstinct.com/2019/04/18/hp-transforms-pc-security-with-ai-driven-hp-suresense/">HP Sure Sense</a>.
This tailor-made security solution involved a complex design for millions of devices worldwide. With our D-Cloud infrastructure in place, Deep Instinct was uniquely positioned to offer a solution with strict update time, availability and scalability to meet the production needs of HP.
As a commercially available product, HP Sure Sense is a self-managed solution. It has no management console because the deployment environment doesn’t have any security officers who would be following events and analyzing the type of events that are occurring.
For HP, this necessitated a high level of confidence in the inference model decision, because there is no course of remediating actions. Deep Instinct’s strong solution’s detection rate and low level of false positives became crucial to the security product’s success.
In delivering the project for HP, a key objective for our team at Deep Instinct was to provide updated services and threat intelligence that would scale to meet the customer’s anticipated increased volume, while also addressing operational and product requirements.
Such requirements included providing a CDN, serverless infrastructure, load balancing, and domain name services (DNS) at the most cost-effective rate. Deep Instinct worked closely with AWS solution architects to define, configure, and design the infrastructure.
Here’s how AWS services were incorporated into HP Sure Sense:
<ul>
<li><a href="https://aws.amazon.com/cloudfront/">Amazon CloudFront</a> is a fast CDN service that securely delivers data, videos, applications, and APIs to customers globally with low latency and high transfer speeds, all within a developer-friendly environment.</li>
<li><a href="https://aws.amazon.com/s3/">Amazon S3</a> is the object storage service containing the raw metadata to support the training cycles.</li>
<li><a href="https://aws.amazon.com/api-gateway/">Amazon API Gateway</a> provides scalable integration with multiple sources.</li>
<li><a href="https://aws.amazon.com/route53/">Amazon Route 53</a> provides a DNS service to support communication.</li>
<li><a href="https://aws.amazon.com/lambda/">AWS Lambda</a> fetches the details and passes them on to the agent via Amazon API Gateway.</li>
<li><a href="https://aws.amazon.com/dynamodb/">Amazon DynamoDB</a> is a NoSQL database that runs completely on serverless infrastructure.</li>
<li><a href="https://aws.amazon.com/ec2/">Amazon EC2</a> hosts all of the computations.</li>
</ul>
This infrastructure was designed with the ability to perform upgrades and updates for the <a href="https://www.deepinstinct.com/">deep learning</a> model infrastructure in a way that would enable significant scaling.
The infrastructure was tested with a high-volume agent simulation, and once it proved successful was ready for implementation with HP Sure Sense.
Summary
With AWS, Deep Instinct’s product architecture was easy to apply with a payment model that scales well for growth. The user interface is supported by a large community of users and a wealth of informational resources, including user manuals and tutorials. This all contributed to the successful completion of the project within management constraints.
To learn more about Deep Instinct’s product architecture, <a href="https://info.deepinstinct.com/mof_deep-instinct-technical-brief">download our technical brief</a>.
A critical module of our production architecture, AWS both supports the development and training of the core deep learning neural network (D-Brain) and supports Deep Instinct’s research and threat intelligence services.
The inference capability of the <a href="https://www.deepinstinct.com/">deep learning neural network</a> necessitates the ability to train on as many data samples as possible to ensure enhanced reliability. When it comes to cybersecurity, quality is key. The more iteration for validating models, the better the performance of the deep learning solution.
&nbsp;
&nbsp;
&nbsp;

How Deep Neural Networks Built on AWS Can Help Predict and Prevent Security Threats

The wave of excitement over artificial intelligence is spreading, and like a tsunami is anticipated to touch every possible surface. And yet as this w

ML-vs-DL.jpg

The wave of excitement over artificial intelligence is spreading, and like a tsunami is anticipated to touch every possible surface. And yet as this wave crashes down, many consumers and companies struggle to identify the subsets of AI, and what these distinctions mean. Decision-makers are still unclear about the difference between ‘machine learning,’ and ‘deep learning’, and the significance that this distinction has in keeping their system’s users and data safe.
<a href="https://www.deepinstinct.com/machine-learning-vs-deep-learning/" target="_blank" rel="noopener noreferrer">Machine learning and deep learning</a> are neither synonyms, nor competing technologies, and yet the variation between them makes all the difference for companies attempting to harness the power of data while trying to determine the best path to success. There are lots of wrong terms and usage, and like any great trend, people end-up following them in the wrong way. In this case, by incorrectly applying terms and mainly by focusing implementation on methods that do not bring any real value and have no technological edge. A good example of this is the use of common open source libraries, which is often done the wrong way or for the wrong domain purpose (we’ll expand on this more, further on).
With all this surrounding confusion regarding machine learning vs deep learning, we bring a clear distinction between the two; we demonstrate their distinctive objectives and the respective purposes to which they lend themselves.
&nbsp;
<h3>Back to the Basics: Defining Deep Learning and Machine Learning</h3>
<a href="https://www.deepinstinct.com/what-is-deep-learning/" target="_blank" rel="noopener noreferrer">Deep learning</a> is a subfield of machine learning, which is itself a subset of AI. Deep Learning is trying to mimic the brain’s behavior in the same way that the human brain learns (by being agnostic to the input) – taking in all the data and learning from it continuously and intuitively.
With the ever-growing availability of data that is increasingly informing many disciplines, deep learning is uniquely disposed to leverage massive volumes of data for more intuitive, accurate, and real-time insights. The training of the deep learning algorithm necessitates scaling hundreds of millions of data samples, and improvement is a case of ‘the more data, the merrier’ (as long as your training data is labeled well, and your dataset is qualified).
Deep learning is unique from machine learning in that it is the first, and currently the only, method that is capable of training directly on raw data. Traditional machine learning requires feature engineering, where a human expert effectively “guides” the machine through the learning process by extracting the features that need to be learnt. As it’s based on human analysis, it’s highly limited and relies solely on specific known use cases. In contrast, deep learning can dive into the raw data of the file, without explicitly being told to pay attention to engineered features and analyzes all the available data.
&nbsp;
<h3>Cybersecurity on Steroids</h3>
In its <a href="https://www.deepinstinct.com/deep-learning-by-deep-instinct/" target="_blank" rel="noopener noreferrer">application to cybersecurity</a>, the deep learning model is able to analyze the millions of different possible files and attack vectors. As the training dataset gets larger and larger, the deep learning algorithms continuously improve. It’s this unique ability to pick up on patterns and nonlinear correlations in the raw data that are too complex for any human or traditional AI to pick up on, which gives deep learning its inherent value.
This ability for in-depth analysis predisposes deep learning to cybersecurity on multiple levels. Firstly, because the entire file is analyzed, it is not limited to selected information that has been extracted. Secondly, aggregating simple non-linear units into a deep network enables the capture of hierarchical dependencies between features which produces empirically better accuracy. Lastly, the abundance of data in which to train with, means that there is great scope to improve accuracy and reliability.
The combination of these analysis features provides numerous valuable outcomes for the future of cybersecurity:
<ul>
<li>Deep learning-based model is able to produce a much higher detection rate and lower false alerts, effectively eliminating alert fatigue and reducing costs.</li>
<li>This is the case even for new and previously unseen cyberattacks. The broad coverage of attack vectors provides verdicts that are more accurate than the best traditional machine learning solutions available.</li>
<li>It provides the ability to scan any type of file statically in pre-execution mode. Unlike detection and response tools that only act when something is already running, which in most cases is too late.</li>
<li>The platform-agnostic solution is able to assess for unlimited types of attack vectors, while it provides protection irrespective of the operating system or the device (be it mobile, endpoint, server, network). Doing so, from a single unified platform.</li>
</ul>
&nbsp;
<h3>Other Applications of Deep Learning:</h3>
Despite the high levels of entry, the adoption of deep learning is accelerating and we see it being applied within particular<a href="https://www.cbinsights.com/research/industries-disrupted-deep-learning/"> commercial sectors</a>:
<ul>
<li>Autonomous cars: The development of image recognition is paving the way for autonomous cars to unlock with facial recognition, alert drivers to potential hazards, and gaze tracking for driver distraction alerts or and driving autonomously. While this technology has been under development for some time, only during the last few years has the pace towards obtaining industry regulation quickened as scientists now have the data to demonstrate that autonomous cars are safer than human drivers. As more and more autonomous cars are operating on the roads and communicating with each other, it will contribute to much more available data. And as we mentioned earlier, the more data the better, as it leads to better accuracy, minimizing the risk and machine false alarms.</li>
<li>Healthcare: Similar to the domain of cybersecurity, the health industry lends itself to deep learning. This is because of the copious amount of information that comes into a medical diagnosis, with tremendous lifesaving results to be gained with even classic machine learning algorithms and implementations. A good example of this is <a href="https://www.researchgate.net/publication/314865626_MobiGuide_a_personalized_and_patient-centric_decision-support_system_and_its_evaluation_in_the_atrial_fibrillation_and_gestational_diabetes_domains">MobiGuide</a> that enables remote patient care, by providing patients with a device that monitors their health, all data is transmitted to a DSS (Decision Supporting System) that is able to analyze all the available data, provides a diagnosis as well as develop a treatment plan. This information is then relayed back both to the patient and doctor for review.</li>
<li>By feeding some of the raw data into deep neural networks without any pre-processing, researchers are getting even greater results. An example of this is <a href="https://medium.com/syncedreview/ai-powered-fda-approved-medical-health-projects-a19aba7c681">IDx-DR</a> a diagnostic system that detects signs of diabetic retinopathy in retinal images. The medical expert system, which has FDA approval can be used without a doctor, to analyze medical images of eyes to determine if treatment is required.</li>
</ul>
 
<h3>And Yet Deep Learning is Not Always the Answer</h3>
While it is true that since 2012 (where a report on the usages of deep learning for state-of-the-art image recognition was published) deep learning has become increasingly commodified, its use has so far been limited to certain fields; of which object recognition is the most persevered application. In the realm of visual object recognition, the machine now outperforms the human being. As training continues based on millions of images, machines are more successful at identifying an object inside images than humans. The most well-known benchmark for humans vs. machine on a respectful dataset is ImageNet which achieved an error rate of 5%, Google superseded this benchmark with a demonstrated 4% error rate, and this has been superseded by Microsoft with a publication demonstrating that their deep learning algorithm achieved an <a href="https://arxiv.org/abs/1512.03385">outstanding 3.57% error</a> on the ImageNet test set.
But this success does not mean that deep learning has been able to successfully enter every field of knowledge. The cost in time and resources involved in building deep learning models entails numerous considerations that would not justify the value of a deep learning model over traditional machine learning models.
<ul>
<li>The domain needs to be an area of knowledge that a traditional machine learning model cannot solve or in which it can reach a limited Considering that machine learning algorithms are much easier to use, if it can adequately do the job, it is not justified to go the extra mile with a deep learning model.</li>
<li>The domain must involve huge amounts of data in which to train. Unlike classical machine learning implementations which rely on certain features, deep learning, necessitates a significant amount of knowledge to build the datasets. So that with the right data and data scientists, you can always expect to increase your recall (until a certain threshold, which would almost always be lower than 100% accuracy) and lower your False Positive rate in correlation to the increase in the size of the datasets.</li>
</ul>
If the input data is flexible, then you should be able to adapt the data to meet the training needs of the fixed size of neurons in the input layers. For example, an image that has too many pixels can be easily reduced by minimizing the size of the image.
When looking to develop a deep learning framework, there are a number of publicly available options, including TensorFlow by Google, PyTorch by Facebook, and CNTK by Microsoft. The availability of these frameworks has significantly boosted research in deep learning, as they enable applications and programs to be implemented directly on these frameworks and experiments to be executed without having to write a single line of low-level code.
However, there are distinct advantages of using a framework dedicated to a particular purpose and objective, instead of public frameworks that are built more for academic educational purposes. Firstly, while a public framework may be designed to have a faster model training time, a framework built for commodification also has an emphasis on a faster prediction time. Secondly, using a public library framework, the agent file would be enormous because there are so many dependencies which it relies on, however, a dedicated framework involves a smaller subset of algorithms (that is more focused to achieving its objectives) enabling the agent to be small enough for commodification.
 
<h3>Beyond the Hyperbole</h3>
With all the buzz around AI, one would be sure that this new field of deep learning will change every aspect of our everyday lives, but the reality is the fully realized potential of deep learning is still a long way off.  This is partially due to the high barrier of entry which will limit the players and the fields in which they interact into image recognition, speech recognition, natural language processing, recommendation systems, and cybersecurity.
Saying that, as deep learning applications continue to expand, companies will ignore these developments at their own peril. The industries that are impacted by deep learning stand to benefit enormously, as the inference model of deep learning continues to supersede human capability, powerfully pushing forward the frontier of technological prowess.
Like any new evolutionary technology, in the coming years undoubtedly, we will find visionaries adapting it and running with it to achieve spectacular implementations, such as analyzing cancer cells and curing other horrible diseases to the result of potentially enabling humans to live longer and safer. Deep learning could be used towards empowering law authorities to effectively fight crime, whether it be detecting fraud or preventing money laundering, by keeping well ahead of criminals. However, deep learning could also be used towards nefarious purposes, either by nation-states or independent hackers, who will try to take advantage of implementing AI to achieve the fulfillment of warped and dangerous ideas. How the future of deep learning will unravel, we’re still yet to tell.
As Andrew Ng, one of the top researches in the Deep Learning field said: “To the AI Community: You have superpowers, and what you build matters. Please use your powers on worthy projects that move the world forward”.

AI Beyond the Semantics: Here's What you Need to Know

The objectives of Artificial Intelligence are to enhance the ability of machines to process copious amounts of data and automate a broad range of task

threat-potential.jpg

The objectives of Artificial Intelligence are to enhance the ability of machines to process copious amounts of data and automate a broad range of tasks. The 1950’s saw the earliest forms of AI, where techniques were developed that enabled computers to mimic human behavior. This gradually evolved into the machine learning of the 1980’s where AI techniques gave computers the ability to learn without being explicitly programmed to do so. More recent years have explored a new subset of machine learning, called ‘Deep Learning’ or Artificial Neural Networks where computers are able to process and assimilate new data, in a similar manner to the human brain.
There are numerous real-world<a href="https://www.deepinstinct.com/what-is-deep-learning/" target="_blank" rel="noopener noreferrer"> applications of Deep Learning</a>. The more commonly known areas are autonomous self-driving cars, image recognition as used by Facebook and Amazon and Natural Language Processing as used by Google and Apple. Yet, for the most part, the field of Deep learning is still in its infancy, where the range of tasks to be transformed by AI are still to be realized, and so are the potential threats.
We explore the expanded threat potential that has been ushered in by Deep Learning technology and how it has further expanded the scope of threats compared to previous computational capabilities. “With great power comes great responsibility” and with that in mind, we finally consider the moral implications.
&nbsp;
<h3>Amplified Capability of AI</h3>
The original motivation towards developing AI technology was to create a system that can perform any given task faster and more accurately than a human. This objective has been met. Human-level performance is no longer the highest standard in which a task can be achieved. For example, the steady progress of AI has made it better at playing chess then even top performing players. On May 11, 1997, DeepBlue an IBM computer beat the world chess champion after a six-game match. The game’s combination of simple rules and multiple challenging problems made chess perfect for such an experiment.
The amplified efficiency of AI means that once a system is trained and deployed, malicious AI can attack a far greater number of devices and networks more quickly and cheaply than a malevolent human actor. Given sufficient computing power, an AI system can complete the one task in many more instances. This increased scalability is demonstrated by Google’s or Facebook’s facial recognition algorithms. If Google’s image net model can classify a million images in less than a few minutes, a human being, no matter how fast, is no competition.
&nbsp;
<h3>The Democratization of AI</h3>
Access to software and new scientific findings is relatively easy. The AI industry has a culture of openness, where papers are published with source codes and AI developments often get reproduced in a matter of weeks, if not days. Thus, in the cat and mouse race between the good guys and the bad guys, much of the time they’re equally advancing the threat landscape in the share of information. For example, both have been known to publish newly discovered malware in the same communal forums. However, it’s not just the knowledge diffusion happening in AI that is potentially expanding cyber threats, programs in robotics and the declining cost of hardware is making technology more accessible to malicious actors.
As AI knowledge and systems become more prolific there is an expectation that it will not only expand the pool of actors who are equipped to carry out an attack but the frequency of committing attacks and the range of possible targets. The combination of knowledge diffusion and scalability means that more malicious actors can enter the field, where once having obtained the now lower threshold of knowledge and resources they can scale up the veracity and frequency of attacks. This makes their motivation and effort far more worthwhile, both in terms of a cost-benefit analysis and in the prioritization of numerous targets, be they based on remuneration or devastation.
&nbsp;
<h3>IBM’s DeepLocker</h3>
A demonstration of this expanded threat potential was made by IBM at the 2018 Black Hat event where they showed the potential capability and impact of deep learning-based ransomware, Deep Locker. The attack comprised a face recognition algorithm that autonomously decided which computer to attack with its encrypted ransomware. The highly targeted and evasive attack tool hides in other applications until it identifies a target victim. Once the unfortunate target is identified through various indicators, the AI algorithm will ‘unlock’ the malware and launch the attack.  The identifying criteria can be any number of attributes including facial feature recognition, audio, location or system-level features.
To test the attack Big Blue IBM hid the ransomware in a video conferencing application where they knew it wasn’t going to be detected and trained the AI model to unlock it based on facial recognition. When the deep neural network recognized the correct facial features of their target in front of their PC, via the webcam, the target’s face became the necessary key to open the payload and lock down their operating system.
The revelation of DeepLocker reverberated strongly across the cybersecurity industry for a couple of reasons. Firstly, it demonstrated the expanded range of triggers in which a ransom attack can be buried, making it difficult to reverse engineer the ransomware attack that is buried within the deep neural network. While there are indeed multiple triggers in which to embed the malicious business logic, the reverse engineer can simply skip this investigative stage to identify how the trigger was used, rather than how it worked. The difference appears niched, but it’s significant. It’s the relative difference for example, of knowing how electricity works and knowing how to just use it.
The second and ultimately more pressing concern was the method, whereby the AI malware was part of the business logic. The real breakthrough was that the attack was AI triggered, rather than triggered by a human being. Marc Stoecklin, Principal RSM and Manager at IBM explained it this way, "the deep convolutional network of the AI model is able to convert the concealed trigger condition itself into a ‘password' or ‘key' that is required to unlock the attack payload." In other words, the AI algorithm was able to not only identify but also trigger the attack against a potential target.
&nbsp;
<h3>Enlarged Psychological Distance</h3>
Cyberattacks have always been characterized by psychological distance and anonymity. The malicious actor never comes face to face with their targets, nor sees the impact of what they have unleashed. Sitting in an enclosed room, opposite a computer throughout an entire attack, there is often a lost sense of the full devastation caused.
AI can also facilitate an even greater degree of psychological distance from the people that are impacted. In the case of an autonomous offensive weapon, such as the terminator, a perpetrator can avoid being present at the scene of the crime and avoid having to see their traumatized victim. The attacker may release on the public numerous terminators in numerous locations around the world, each with variant tasks and each with their own violent capabilities.
Autonomous AI-driven offensive weapons are to be found well beyond the scope of science fiction.  AI-based autonomous weaponry is currently being developed in Russia, USA, UK, China, and Israel among other locations. In the case of an AI-driven weapon, there is no need to control and supervise it, an attacker could launch as many attacks as they want with just a hand full of people, be more selective in their target and more devastating in their impact. Under these circumstances it would be possible to have, let’s say, 10 people, launching 50 million weapons, causing mass destruction on the same level as a large nuclear explosion.
&nbsp;
<h3>The Moral Implication</h3>
The knowledge of how to design and implement AI systems can be applied to both civilian and military purposes and likewise towards beneficial and harmful ends. In the same way that human intelligence can be used towards positive, benign or detrimental purposes, so can artificial intelligence. Many tasks that lend themselves to be automated, have this multifarious aspect. Consider systems such as Metasploit that examine software for vulnerabilities, this knowledge can be used to either rectify or exploit the vulnerability. Likewise, the difference between a drone used for delivering packages and a drone used to deliver explosives is not all that different.
How we, as a global community, choose to expand the AI frontier will become critical. Even when knowledge is pursued for wholesome purposes, there is no guarantee that its end application will maintain that same wholesome outcome. There have been many to call for regulatory control to help reign in its expanding development. Including Elon Musk, founder of Tesla who said “we need to be careful with AI. Potentially more dangerous than nukes… I’m increasingly inclined to think there should be some regulatory oversight (of AI), maybe at the national and international level”.
However, others feel that attempts to regulate ‘AI’ in general would be misguided. After all, how do you regulate something that doesn’t have any one clear definition? And where the risks and considerations are vastly different depending on the domain.
Possessing undeniably good and bad potential in equal parts, there is ultimately no way to control the global exploration and<a href="https://www.deepinstinct.com/what-is-deep-learning/" target="_blank" rel="noopener noreferrer"> application of AI</a>. Its threats have to be treated as a potential danger, in the same pragmatic approach that you protect your PC with antivirus software to help block dangers. The critical factor is having a defense system in place that is more technologically advanced than the impending threat. Enterprises most susceptible to an advanced AI attack must have in place the necessary technology that is able to mitigate the developing risks of AI.
&nbsp;
&nbsp;
To learn more about Deep Learning, how it works and how it differentiates from machine learning, download the Ebook <a href="https://info.deepinstinct.com/deep-learning-for-dummies" target="_blank" rel="noopener noreferrer">Deep Learning for Dummies</a>. 
<a href="https://info.deepinstinct.com/deep-learning-for-dummies" target="_blank" rel="noopener noreferrer"><img class="alignnone wp-image-3003" src="https://www.deepinstinct.com/wp-content/uploads/2018/07/dummies-blog-banner-450x113.png" alt="" width="647" height="164" /></a>

The Advanced Threat Potential of Deep Learning

In recent years, deep learning and machine learning have gained traction in so many areas that have a direct positive effect on our lives as well as c

blog3-copy-12.jpg

In recent years, <a href="https://www.deepinstinct.com/2019/04/16/applications-of-deep-learning/" target="_blank" rel="noopener noreferrer">deep learning</a> and machine learning have gained traction in so many areas that have a direct positive effect on our lives as well as complex tasks such as computer vision (image recognition), machine translation, and natural language processing. And with like so many other technologies that are changing our lives for good, it has the destructive potential to change it for bad, there is no reason why it won’t also be used for malicious activities as well. Up until now, we haven’t seen the use of AI for malicious activities in cybersecurity due to the high costs, lack of skills and the tools available. But just like any other technology, it’s a matter of time before it happens in cybersecurity.
AI vs. AI
Think about what would happen when attackers start using the power of deep learning and machine learning for their advantage?
That being said, currently, the use of AI for attackers is mainly being used in academia and not in practical attacks.
But there’s a lot of talk in the industry about attackers using AI in their malicious efforts, and defenders using AI as a defense technology. And we’re here to make sense of this.
There are three types of attacks in which an attacker can use AI:
<ol>
<li>AI-boosted/based cyber-attacks – In this case, the malware operates AI algorithms as an integral part of its business logic. For example, using AI based anomaly detection algorithms to indicate irregular user and system activity patterns. Unusual patterns can lead to different malware behavior, increased or decreased evasion and stealth configurations, and communication times. Situational awareness is implemented in malware for a long time, but AI can offer much more accurate and adaptive approaches.</li>
</ol>
<a href="https://securityintelligence.com/deeplocker-how-ai-can-power-a-stealthy-new-breed-of-malware/" target="_blank" rel="noopener noreferrer">An interesting use case</a> can be found in DeepLocker, presented by IBM Security researches in Black Hat USA 2018. DeepLocker is encrypted ransomware which autonomously decides which computer to attack based on a face recognition algorithm - meaning only when the target is recognized by the camera (after using face recognition techniques to identify) the attack takes place.
There are other hypothetical use cases, which might be a part of malware business logic. Consider “Anti-VM”, for instance. Sophisticated malware tends to check if it runs on a virtual machine (VM), to avoid operating its malicious activities on sandboxes, which will reveal the file is malicious, or to avoid being analyzed by a security researcher, which might reveal how it works. In-order to assist their Anti-VM efforts, malware writers can train a VM environment classifier, that would get environment details (e.g., registry keys, loaded drivers, etc.) as features and understand whether the host the malware is running on is a VM or not. Moreover, such a model can resolve some of the difficulties malware have when they run on cloud hosts, which are also VMs, but not security research-oriented VMs, increasing the malware spread.
&nbsp;
<ol start="2">
<li>AI-based attack infrastructure and frameworks – in this case, the malicious code and malware running on the victim’s machine do not include AI algorithms, however, AI is used elsewhere in the attackers' environment and infrastructure – on the server-side, in the malware creation process etc.</li>
</ol>
For instance, info-stealer malware uploads a lot of personal information to the C&amp;C server, which then runs an NLP algorithm to cluster and classify parts of the information as “interesting” (credit card numbers, passwords, confidential documents, etc.).
Another example for this would be <a href="https://en.wikipedia.org/wiki/ICloud_leaks_of_celebrity_photos" target="_blank" rel="noopener noreferrer">#TheFappening</a> attack, where celebrity photos stored on iCloud were leaked. An attack like this one could have taken place in a much larger scale if it was an AI facilitated attack. For instance, computer vision machine-based algorithms could be used to review millions of pictures and identify which of them contains celebrities and then expose only the matching ones, similar to the ones leaked in #TheFappening.
Another example for an AI-facilitated cyber-attack can be a spear-phishing attack, as described in the report: <a href="https://arxiv.org/ftp/arxiv/papers/1802/1802.07228.pdf" target="_blank" rel="noopener noreferrer">The Malicious Use of Artificial Intelligence: Forecasting, Prevention, and Mitigation</a>. In phishing techniques, the target is “fooled” by a superficial trustworthy façade tempting them to expose sensitive information/money, whereas a spear-phishing attack involves collecting and using the information specifically relevant to the target, causing the façade to look even more trustworthy and more relevant. The most advanced spear-phishing attacks require a significant amount of skilled labor, as the attacker must identify suitably high-value targets, research these targets’ social and professional networks, and then generate messages that are plausible within this context. Using AI – and specifically generative NLP models, this can be done at a much larger scale, and in an autonomous way.
&nbsp;
<ol start="3">
<li>Adversarial attacks – In this case, we use “malicious” AI algorithms to subvert the functionality of “benign” AI algorithms. This is done by using the same algorithms and techniques used in traditional machine learning, but this time it’s used to “break” or “reverse-engineer” the algorithm(s) of security products. For instance, <a href="https://en.wikipedia.org/wiki/Stochastic_gradient_descent" target="_blank" rel="noopener noreferrer">Stochastic Gradient Descent</a> which is a technique used to train deep learning models that can be used by adversaries to generate samples that are misclassified by machine learning or deep learning algorithms.</li>
</ol>
<a href="http://techpolicylab.org/new-adversarial-machine-learning/" target="_blank" rel="noopener noreferrer">One example of adversarial learning</a> is placing a sticker in a strategic position on a stop sign, causing it to be miss-classified by an image recognition street sign classifier as a speed limit sign. Another example of this attack is <a href="http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.187.8026&amp;rep=rep1&amp;type=pdf" target="_blank" rel="noopener noreferrer">injecting malicious data streams into benign traffic</a> in order to cause an anomaly detection-based network intrusion detection system (NIDS) to block legitimate traffic effectively causing Distributed Denial of Service attacks (DDoS).
Such attacking techniques have been developed by researchers against computer vision algorithms, NLP and malware classifiers.
Brace Yourself!
We believe the AI vs. AI trend will continue to increase and cross the border from academic POCs to actual full-scale attacks as computing powers (GPUs) and deep learning algorithms become more and more available to the public.
In order to have the best defense, you need to know how attackers operate. <a href="https://www.deepinstinct.com/machine-learning-vs-deep-learning/">Machine learning and deep learning</a> experts need to be familiar with these techniques in order to build a robust system against them. Keep up with the <a href="https://www.deepinstinct.com/what-is-deep-learning/">AI threat potential</a> landscape and watch this <a href="https://info.deepinstinct.com/on-demand-webinar-current-trends-and-the-evolution-of-modern-cyber-threats" target="_blank" rel="noopener noreferrer">on-demand webinar</a> on the evolution of modern cyber threats.
&nbsp;
<a href="https://info.deepinstinct.com/on-demand-webinar-current-trends-and-the-evolution-of-modern-cyber-threats" target="_blank" rel="noopener noreferrer"><img class="alignnone wp-image-6026" src="https://images.contentstack.io/v3/assets/blt1ec077b6b53d6b3e/blt90117e419750f7be/611a7eda75b34916435392d0/webinar1.png" alt="" width="560" height="329" /></a>

How AI can be used for Malicious Purposes

Unit 221B Validates >99% Zero-Day and <0.1% False Positive Threat Prevention Accuracy