Hello everyone. I'm Dineh Raja Sharan. I'm honored to be here at Con 42 to talk about the future of cybersecurity. As someone deeply involved in both academia and industry, I've seen firsthand how the cybersecurity landscape is evolving. And how AI is not just a tool, but a game changer. Over the next 30 minutes, I will walk you through the transformation from reactive to predictive cybersecurity, focusing on how AI is enabling organizations to stay ahead of ever more sophisticated cyber security threats. Let's dive in. To app appreciate the urgency of this transformation. Consider this, the global cost of cyber crime is expected to reach 10.5 trillion annually by 2025. That's up from $3 trillion just a decade ago. This explosion is driven by everything from ransomware to phishing, data theft nation state attacks. The good news is we have tools today that can help us fight back. AI is one such tool. MIT's a AI two platform, for example, detects. 85% of attacks and a financial firm using AI saw a 60% drop in false positives. These are in incremental improvements. They are breakthroughs. So why isn't traditional security enough anymore? First, it's reactive. It only kicks in once damage is already done. Second, the sheer volume of data in modern networks overwhelmed static rule-based systems. Third, traditional systems depend on known attack vectors or signatures. That means if an attacker creates a new tactic. A zero day your defense is blank. This is like using a map from the past to navigate a rapidly shifting landscape. AI allows us to move beyond that limitation by learning from patterns, not just pass emails. AI changes the game by enabling real time. Dynamic threat detection. It starts with comprehensive data collection. Everything from firewall logs to user behavior. Then it pre-process the data cleaning and structuring it from analysis. Next AI models, the large language models dive into the data. Identifying subtle patterns or anomalies that humans might miss. Finally, the system provides security teams with actionable insights, sometimes even triggered automated containment protocols. The result is faster detection, better prioritization, and more efficient response. And how do we go about this? So there are three foundational techniques at play here, anomaly detection, pattern recognition, and real-time threat intelligence. Anomaly detection uses historical data. To flag behaviors that deviate from the norm. Pattern recognition identifies complex combinations of events that may signal a threat, and the real time threat intelligence integrates internal data with global threat feeds to assess risk asset intervals. And then they act together. Together. These techniques enable systems to evolve and adapt just as quickly as attackers do. And then let's. Zoom into the three models, what we discussed in the previous slides, ly detection, pattern recognition, real time threat intelligence. Let's consider the example of a supervised learning. Supervised learning uses labeled data sets like previous. Phishing attacks leveraging the previous phish phishing attacks. You can predict similar future attacks. On the other hand, unsupervised models are more exploratory. They cluster data, they find outliers. They're perfect for spotting new kinds of threats. Then there's deep learning which shines in high volume, high complexity environments like cloud infrastructure. These models, especially the the nns and CNNs of the world, are able to process logs, sensor data, user actions in a nuanced way, identifying threats no human code. Normally detection is not a one time setup. First, it builds a baseline of normal operations. Who locks in when, what actions are taken. Then in real time, it flags deviations. For instance, if an employee suddenly accesses an entire database at 3:00 AM which is not a normal pattern, the system notices. More importantly, these AI systems evolve. They adapt their baseline to reflect changes in normal behavior, minimizing false positives over time. They get smash smarter, not just faster click. Let's look at three case studies, for example. Let's take a financial institute, let's say a major bank integrated AI for fraud detection, and saw not only fewer false positives, but a 50% increase in new fraud types detected. An e-commerce chain uses AI to mitigate DDoS attacks, reducing impact by 91st, 95%. In healthcare, AI identified subtle patient records, access patterns that flagged insider threats. These stories show how AI not only helps protect, but also optimizes how teams operate, freeing them to focus on critical tasks. Speaking of the pattern recognition from the previous slides pattern recognition is where AI truly shines. Attackers don't always follow the same path, but they are often common sequences like login anomalies followed by privilege escalation. AI mines historical logs to find these attack fingerprints. With the graph based analysis, AI can even detect coordinated campaigns across distributed systems. This transforms how we investigate threats from reactive foreign six two proactive hunting, the real time threat intelligence part. AI also uses, I said, ears across the network 24 7. It correlates activities across endpoints servers, cloud apps to spot risks early. More importantly, it reacts in real time. Some systems can isolate infected machines. Automatically. But by precious time and by sharing threat intelligence across organizations, we are creating a more resilient digital ecosystem. Studies show organizations using threat intelligence reduce breach cost by over 2 million. However it's not all rosy. AI requires data. Specifically it requires clean label data which often could become a bottleneck. Poor data leads to poor models, and then there's a trust gap. If a model flags a threat can be explained why this black box nature can be a compliance challenge also. False positives still happen. And then if frequent, and then if frequent, they desensitize analysts, a phenomenon known as alert fatigue, lots of alerts lots of false positives. Lastly, ethical concerns around surveillance data usage must be addressed thoughtfully. The future though. Looking ahead, if you see explainable AI models that just justify their decisions adversarial machine learnings to defend against AI attacks, federated learnings for privacy, preserving analysis and analytics and fully autonomous security systems. So these. Models, advances are not decades away. They're already being tested in research labs, in early stage developments. Right now, there are several startups that have been spinning up with the same models. So in summary, AI isn't just the future of cybersecurity, which is going to happen like 10 10 years from now. It is happening right now. It's present. But it's real value lies in collaboration between humans and machines. Between ethics and innovation. If you build AI thoughtfully the result will be stronger, smarter, and more sustainable security systems that adapt and scale without increasing without increasing increasingly digital work. Thank you.