Machine learning is a branch of artificial intelligence that enables computers to mimic human cognition through empirical learning and adaptive techniques. It is characterized by learning based on experience and patterns rather than on inferences (causes and outcomes). At present, deep learning in machine learning has been able to build a pattern recognition model independently, without having to rely on humans to build models.
Traditional network security technologies are difficult to detect a new generation of malware and cyber attacks that evolve over time. ML-based dynamic network security solutions can leverage previous cyberattack data to address new but similar risks. Using AI to enhance network security can provide additional protection for user systems, such as automating complex processes to detect attacks and react to violations.
As the pattern recognition model becomes more effective in detecting cybersecurity threats, hackers will study the working and learning mechanisms of the underlying model, find effective ways to confuse the model to avoid the identification of the model, and hope to establish itself as an attacker. AI and machine learning tools to launch attacks.
Below, the author will share with the prince how the attacker will use AI to achieve the goal.
Malware escapeMost malware is generated manually, and attackers write scripts to generate computer viruses and Trojan horses, and use rootkits, password crawlers, and other tools to assist with distribution and execution.
Can this process speed up? Can machine learning help create malware?
Machine learning methods are an effective tool for detecting malicious executables. Learning from data retrieved from malware samples (such as header fields, instruction sequences, or even raw bytes) can create models that distinguish between benign and malware. However, analyzing security intelligence can reveal that machine learning and deep neural networks are confusing by evasive attacks (also known as confrontational samples).
In 2017, the first example of publicly using machine learning to create malware was presented in the paper "GeneraTIng Adversarial Malware Examples for Black-Box Attacks Based on GAN." Malware authors often do not have access to the detailed structure and parameters of the machine learning model used by the malware detection system, so they can only perform black box attacks. The paper reveals how to generate anti-malware samples by building a genetic algorithm (generaTIve adversarial network, GAN) that bypasses machine learning-based black box detection systems.
If the AI ​​of the network security enterprise can learn to identify potential malware, then the "hacker AI" can make decisions by observing the anti-malware AI and use that knowledge to develop "minimally detected" malware. At the 2017 DEFCON conference, security company Endgame revealed how to use Elon Musk's OpenAI framework to generate custom malware, and the malware created could not be detected by the security engine. Endgame's research is based on seemingly malicious binary files. By changing some of the code, the changed code can evade detection by the antivirus engine.
In March of this year, the paper "Adversarial Malware Binaries: Evading Deep Learning for Malware DetecTIon in Executables" proposed a gradient-based attack by investigating the vulnerabilities of using deep networks to learn malware detection methods from raw bytes: Small changes in data can lead to misclassification at the time of testing, so by changing a small number of specific bytes at the end of each malware sample, you can evade security detection while retaining its intrusion capabilities. The results show that modifying less than 1% of the bytes, against the malware binary, can avoid security detection with high probability.
2. Advanced spear phishing attackA more obvious application against machine learning is the use of text-to-speech, speech recognition and natural language processing algorithms in intelligent social engineering to teach software's e-mail writing style through time recurrent neural networks, making it authentic and credible. Sex is enhanced. So in theory, phishing emails may become more complex and credible.
In the 2017 forecast of McAfee Labs, criminals will increasingly use machine learning to analyze large numbers of stolen records to identify potential victims and build detailed phishing electronics that can more effectively target these people. mail.
In addition, at the 2016 US Black Hat Conference, John Seymour and Philip Tully published a paper entitled "Weaponzing data secience for social engineering: automated E2E spear phishing on Twitter", proposing a time recurrent neural network SNAP_R, learning how to Specific users publish phishing posts. Here, spear phishing uses the posts posted by users as training test data. According to the target users (including post or post users), the dynamic seeding of the topics in the timeline posts makes the phishing posts more likely to be clicked. . By testing on the Twitter social platform, it was found that the phishing posts tailored for users have the highest click-through rate in large-scale phishing attacks ever reported.
3. Use AI to beat the verification codeAt present, the distinction between people and machines is mainly based on "Completely Automated Public Turing test to tell Computers and Humans Apart (CAPTCHA)", commonly known as verification code to prevent people from using automated robots. Set up a fake account on the website. When logging into a website, users must prove that they are human by solving visual problems, and this requires identifying letters, numbers, symbols, or objects that are distorted or animated in some way. The reCAPTCHA project is a system developed by Carnegie Mellon University. The main purpose is to use CAPTCHA technology to help digitize the classics. This project will be scanned by books and cannot be accurately identified by Optic Character Recognition (OCR). The text is displayed in the CAPTCHA problem, allowing humans to recognize these words with the human brain when answering CAPTCHA questions.
As early as 2012, researchers Claudia Cruz, Fernando Uceda, and Leobardo Reyes released an example of a machine learning security attack. They used the support vector machine (SVM) to crack the image runtime reCAPTCHA with an accuracy of 82%. As a result, all the verification code mechanisms have been targeted for security improvements. In the face of these new verification code systems, researchers have begun to try to use depth. Learning techniques are broken.
Vicarious has been developing algorithms for the Procursive Cortical Network (RCN), which aims to identify objects by analyzing the pixels in the image to see if they match the contour of the object. In 2013, Vicarious announced that it has cracked the text-based captcha test used by Google, Yahoo, PayPal and Captcha.com with an accuracy rate of 90%. In the standard reCAPTCHA test, the software can successfully solve two-thirds of the verification problems. In the robot detection system test, the success rate of Yahoo verification code is 57.4%, and the success rate of PayPal is 57.1%.
Last year's "I am a robot" study at BlackHat revealed how researchers cracked the latest semantic image CAPTCHA and compared various machine learning algorithms.
4. Bypass the security detection phishing page"Cracking Classifiers for Evasion: A Case Study on the Google's Phishing Pages Filter" pointed out that the phishing web classifier in Google is obtained through machine learning training, and the attacker uses reverse engineering technology to obtain part of the information of the classifier. The generated new phishing page can bypass Google’s phishing web classifier with a 100% success rate. The early development of the classifier is of a research nature, and its security has not received the attention it deserves when deployed in a client environment.
The case of researching the client classifier security challenge is the Google's phishing pages filter (GPPF) deployed on the Chrome browser with more than one billion users. The new attack method for the client classifier is called Crack for the classifier. Successfully crack the GPPF classification model, and obtain sufficient knowledge (including classification algorithms, scoring rules and features, etc.) to conduct effective evasion attacks. The attacker can obtain 84.8% of the scoring rules through reverse engineering, which covers most of the high-weight rules. Based on these cracking information, two evasive attacks against GPPF were implemented. After testing on 100 real phishing pages, it was found that all phishing pages (100%) can easily bypass GPPF detection. Studies have shown that existing client classifiers are vulnerable to classifier targeted attacks.
5. Let the machine learning engine "poison"A simpler and more effective use of AI is to "poison" the machine learning engine used to detect malware, making it ineffective, as criminals have done to anti-virus engines in the past. The machine learning model needs to learn from the input data. If the data pool is "poisoned", the output will also be "poisoned." Deep neural network training requires a lot of computing resources, so many users train in the cloud or rely on pre-trained models to identify and fine-tune specific tasks. Researchers at the University of New York presented a vulnerability in externally trained neural networks in the paper "BadNets: Identifying Vulnerabilities in the Machine Learning Model Supply Chain": opponents can generate a malicious training network (backdoor neural network or BadNets) while The effectiveness of BadNets attacks is demonstrated in the MNIST digital identification and traffic sign detection mission.
Hackers are increasingly using AI vulnerabilities to build "anti-samples" to evade attacks. The current countermeasures are mainly: using game theory or probabilistic models to predict attack strategies to construct more robust classifiers, using multiple classifiers. The system increases the difficulty of avoiding, and optimizes the feature selection to make the average distribution of features. More AI attack coping methods are still being explored.
76-200Kva Diesel Generator,Volvo Generator,Generator Set,Portable Diesel Generator
Shanghai Kosta Electric Co., Ltd. , https://www.ksdgenerator.com