To remove uncertainty from the topic of the world of artificial intelligence (AI) and cybersecurity, Votiro Founder and CTO Aviv Grafi sits down with Eric Avigdor, VP of Products. This series is his three-part “Coffee Talk” series. Explore different aspects of AI and its impact on cybersecurity.
In this first chat, Aviv and Eric introduce the fundamentals of AI, Machine Learning (ML), and Large Language Models (LLM) and discuss the huge opportunities for rapid adoption of generative AI. The two distinguish well-known generative AIs for consumers such as ChatGPT and BARD from generative AIs used for security such as SOC assistants to ensure data integrity and privacy, Enhanced security analysis. We also discuss generative AI used in the enterprise, including training AI models with enterprise data, improving digital business processes, and enhancing business analytics for better overall decision-making.
It’s clear that AI is here to stay, for better or worse. Watch the episode and read the recap below.
The discussion then turns to cybercrime. Cybercrime used to be the domain of highly specialized attackers with significant skill sets and know-how. more than this. AI has become a commodity that anyone can use to commit cybercrime. The Votiro team describes the availability of Cyberware-as-a-Service. In this service, attackers create malware and sell it on the dark web to those who want to wreak havoc on another entity. It also explores the concept of Cybercrime-as-a-Service, where hackers offer malicious services to anyone willing to pay.
For example, AI-powered Ransomware-as-a-Service is being sold on the dark web. For the right price, you can get the best service package deals that include ransomware installation techniques, methods of raising funds, and services that facilitate payments from victims.
Ransomware is the most prevalent, with 24% of breaches involving organizations of all sizes, but it is not the only option available to malicious actors. Buy an AI-powered botnet to bring down his competitors’ websites, or use AI-powered phishing emails to trap even the smartest employees. For example, ChatGPT can help you craft the perfect phishing email without any grammatical or spelling mistakes that would raise suspicion. It also helps you generate perfectly worded ransom note in 28 different languages with just one click.
Graf and Avigdor then draw attention to the dangers of over-reliance on AI. AI and ML can accomplish tasks we never thought possible, but two security experts point out that their effectiveness depends on the data set. Training an AI model requires a large amount of data. If the data is not comprehensive enough, the results will be poor. For example, if a model is trained with insufficient data to recognize phishing emails, it can produce many false positives. And even if you have perfect data that allows your model to detect phishing emails with known parameters, what about zero-day threats and as-yet-unknown threats?
Worse is the concept of tainted data. The expert duo discuss the phenomenon of using natural language to trick learning models. For example, if you tell the model that the earth is flat, that input becomes a fact. So the next time someone asks a question, the poisoned model will respond accordingly.
The same is true for organizations looking to train AI models to improve business or product efficiency. The first step is to collect a huge dataset and train a model to make decisions based on that data. But what if one of the files used in the dataset is tainted, i.e. corrupted or injected with malware? These tainted data files create tainted models. and, possibly maliciously, cause the organization to make the wrong decisions. Data repositories can also accidentally or maliciously contain privacy-related data such as credit card information or HIPAA-related data, all of which can ultimately be misused.
Votiro’s Revolutionary Zero Trust Approach to Content Security Helps Enterprises Deliver Secure Content and Feature files are sent to users and applications no matter where or how the file is entered. Unlike detection-based security solutions that scan for suspicious elements and block some malicious files, Votiro’s technology singles out only the safe elements of each file, ensuring that every file entering your organization is safe. , to ensure that it is free of threats and other harmful substances that may adversely affect the organization. data model.
Grafi and Avigdor have spent the past few years testing thousands of macros, both malicious and benign, to ensure their technology can confidently identify safe macros with the highest accuracy. We revealed how we trained our own AI model on files. These ask the viewer where the organization’s files reside. Files may be downloaded by employees, uploaded by partners, or stored anywhere on the cloud he infrastructure. Votiro gives organizations peace of mind that their pipeline is secure. It’s clean and safe to use because it’s populated with AI models. And it’s a win-win for everyone.
Contact us today to learn more about Votiro and how we can help your organization build a secure foundation. If you’re ready to try Votiro for yourself, start your 30-day free trial today.
*** This is Votiro’s Security Bloggers Network syndicated blog created by Votiro. Read the original post: https://votiro.com/blog/unveiling-the-power-and-pitfalls-of-ai-in-cybersecurity-a-coffee-talk-series/
