2025 Cybersecurity and AI: Predicting the Unpredictable – Predict 2025

Hi everyone. It is an honor to be here today. My name is Caroline Wong, and I'm currently a director of cybersecurity at Teradata.

Throughout my nearly two decade journey in this field, I've had the privilege of working across so many domains from GRC to software security to product innovation. Now at Teradata, my focus is on ensuring the resilience of our systems and safeguarding the sensitive data that fuels transformative insights for organizations worldwide. My career has taken me through roles at leading companies like eBay, Zynga, Symantec, and Cobalt, where I've built and scaled security programs, led global teams and championed innovative approaches to cybersecurity challenges.

I'm passionate about translating complex security concepts into actionable strategies. This is reflected in my book Security Metrics of Beginner's Guide, which was inducted into the cybersecurity Canon Hall of Fame, and my work as a LinkedIn learning instructor, where I empower professionals with the skills to navigate the rapidly evolving security landscape. Today, I'm super excited to share my thoughts on cybersecurity and AI predictions for 2025, drawing on lessons that I've learned from both successes and challenges throughout my career.

Let's explore how we can continue to innovate and adapt to stay ahead in this always changing field. Let's dive in. Prediction number one.

This was kind of a bummer, but I really think that in 2025, humans are gonna end up doing 40% of the work that AI is supposed to do. AI is often marketed as a magical solution, capable of replacing human effort entirely. But in practice, the story's a little bit different.

Real world implementation often falls short because of challenges like poor data, quality, integration issues, and specific domain complexities. Even when AI does provide great insights and recommendations, it faces what is called the last mile problem. AI does an okay job at analysis and prediction, but it takes human validation, interpretation, and action to bridge the gap between analysis and actual impactful results.

I think that unfortunately in 2024, there's been so much investment in ai, and AI is really expected to do so much. I think in 2025, those expectations are gonna fall short. Organizations really should continue to rely on humans to oversee AI decisions, especially in high stakes areas like healthcare, finance, cybersecurity, where trust is key.

Sadly, AI isn't perfect. It can't self-diagnose bias, it can't correct unexpected behavior, and human invention is gonna be absolutely needed to monitor those outputs, catch errors, and make sure that the systems work as they are intended. It would be nice to have full automation in some business processes, but building AI systems that are actually capable of end-to-end automation requires extraordinary resources, time, money, and expertise.

And even then, results are falling short when it comes to dynamic unpredictable scenarios. Another broader societal implication is that I think that there are gonna be workers and even organizations that resist full automation in order to preserve jobs, in order to support historical cultural practices. Um, and even when AI works really well, people are still gonna want some human assurance.

Um, whether we're talking about customer facing rules or business critical decisions, fairness and accuracy and increasingly empathy are still going to be much more trusted when humans are involved. So this prediction is a little bit of a bummer, but the next one is really exciting because I think that in 2025, kids are gonna use AI to solve real world problems. So AI isn't just transforming industries, it's empowering young minds to think bigger and achieve more than we ever imagined.

It starts with the democratization of tools. AI technologies like fat, GPT image generators and coding assistance are now accessible to kids. These tools break down barriers and allow children to experiment with ideas and create solutions that before might have required months or even years of experience or specialized knowledge.

Today's kids are not just dreaming about the future, they're actually building it. They're going to use AI to develop games and applications. They're also gonna use it to address environmental issues and tackle community challenges like waste management or energy efficiency.

AI is really fun for kids. It brings STEM education to life in remarkable ways. Instead of learning abstract concepts in isolation, kids can use AI to engage with hands-on application.

Imagine a middle schooler training a model to analyze air quality or developing a chatbot to support mental health in their local community. Programs like AI for kids are fostering this practical engagement, encouraging students to take on global challenges like climate change and healthcare accessibility. What makes this even more exciting is that kids see the world differently.

They have so much curiosity and creativity when it comes to finding unconventional solutions. Older generations may have labeled some things as insurmountable, but kids don't think that way, and thanks to the internet and computing, they're not as limited by geography. So kids from different regions and backgrounds can collaborate virtually to adjust shared challenges like access to clean water or improving education equity.

So kids are really the future of problem solving, and AI is gonna play a big role here. The next prediction is a strange one. I predict that in 2025, 80% of single people between the ages of 18 and 58, we'll have an AI boyfriend or girlfriend.

So what makes AI companions so compelling is their adaptability and AI companions are coming up in popularity at an exact time when human loneliness and isolation is more pervasive than it ever was before. An AI companion can be customizable. They can learn from a user's preference, they can create deeply personal and tailored interactions.

At this point in time, each of us actually interacts with so many of our real human relationships via technology. And so if a person's talking to an AI companion, you can actually develop feelings as if the AI understands you in ways that other people might not. This is really just the next stage and a broad cultural shift that's been happening over the past couple of decades.

It is in addition to the way that online dating and social media have already reshaped the way that we operate and think about our actual relationships. AI relationships are going to become a natural extension as that for many people, the emotional and the psychological bond with an AI partner is going to feel just as real as one with a human being. The line between a virtual and a real relationship is blurring, and it also gives us cause to rethink what intimacy and connection, uh, truly mean.

Recently at the World Health Organization recognized loneliness as a global health crisis, and so for some of those folks feeling really isolated, an AI partner could actually provide emotional scaffolding that they need in order to feel seen and valued and supported. Of course, there will be challenges. It's gonna be weird if somebody feels like they like their AI relationship more than their traditional real life partnership.

It's also gonna shift societal expectations about love and about intimacy. There are ethical considerations. What happens to humans if we begin to depend too heavily on AI for emotional support?

How is that gonna stop us growing as individuals? How is that gonna prohibit us from forming meaningful actual human relationships? As AI companions become increasingly realistic, there's gonna be questions that need to be answered about things like consent, manipulation and the authenticity of those emotional bonds.

Ideally, I think that there's a future that's possible where AI doesn't replace human connection, but actually enhan enhances it. Maybe AI can help to teach us how to be stronger communicators and more empathetic partners. That's AI and, and artificial boyfriends and girlfriends, uh, which is a weird one, and it actually, it, it kind of leads nicely into the next prediction, which is that I predict that impersonation attacks will increase my 500%.

Now, this is a very dramatic rise, and I think that this is actually pretty conservative because AI is becoming a force multiplier for cyber criminals. AI is revolutionizing your basic phishing attacks. They're becoming hyper personalized.

Attackers can use tools to do things like get information off of your social media, your LinkedIn, your Facebook, your public posts on online forums, and they can use that information in combination with AI to generate emails and messages that are so tailored to your actual life. They can reference your kids' school. They can reference where you went on vacation recently.

They can reference your recent project at work, and this is gonna help them bypass any initial suspicion. AI also provides attackers with an entirely different level of scale and speed. For years now, we've had voice and video deep figs that have been so realistic, they could pretend to be your boss calling you with an urgent request.

CEO fraud is already a billion dollar problem, and it's gonna be increasingly challenging to tell the difference between a real request and an impersonation attempt. One of the things about chatbots and large language models in particular is that in the past, hackers were often limited by their knowledge and ability to use, uh, the native language of their victims. But today, AI generated content can be made so linguistically accurate and it can often be indistinguishable from human created messages.

Um, one of the big takeaways here is to watch out what you're posting on social media. Just keep in mind that that is information that attackers can use in order to personalize spear phishing campaigns. This, of course, is not just a technical challenge more than anything, it's a psychological one.

AI can model human behavior, and for decades, hackers have been exploiting human emotions like fear, urgency, and curiosity to push their victims into making quick decisions. Um, and now attackers can really target their campaigns, um, identifying specific, specific victims, uh, based on their demographics, their profession, um, and their interests. The next topic that I wanna share has to do with hacktivism.

I predict that in 2025, hacktivists will intentionally introduce bias into mainstream AI models. So fundamentally, the way that AI learns is from patterns in the data that it's trained on. And so if that data is biased, whether it's intentional or not, the AI will reflect and even amplify those biases.

Biases. Imagine a scenario where someone is feeding an AI system, hiring data where men were historically favored for leadership roles. The AI upon observing this pattern could perpetuate it inadvertently reinforcing inequality bias and ai.

There's, it's just there. It's there. And now it's not only a byproduct, it's a vulnerability.

Activists can intentionally introduced bias into models because these folks, they're driven by ideology, they're driven by the desire to exploit weaknesses, and this is a way for them to advance their agenda or to make symbolic statements. Already, there seems to be confusion sometimes as to whether or not something you look up on the internet is fact or potentially fiction. Similarly, there are gonna be plenty of people who assume that AI always tells the truth.

And so if hacktivists or other groups are going to be biasing the data in a way so that they're advancing their agendas, um, it could create chaos. It could undermine public trust, it could actually change the way that people think about facts. There are few different ways that this could happen.

One method is via data poisoning attacks. So hacktivists can infiltrate the data sets that are used to train the AI models subtly altering them to introduce skewed or harmful or otherwise inaccurate biases. For example, imagine a hacktivist decides to corrupt a dataset that's used for credit scoring that could lead to unfair lending practices, which might disproportionately affect certain groups.

Another approach is to fine tune exploits. So many AI applications used pre-trained models, and these can be accessed and altered by attackers. Hacktivists may be able to tweak parameters and therefore embed biases that affect everything from hiring algorithms to search engine results.

And then we've got prompt injection. Uh, this tactic targets generative AI systems in particular. So by crafting specific inputs, activists can corrupt outputs in real time spreading biased, misleading or harmful content instantly.

Finally, I think this might actually be my second to last prediction, so I've gotta, I've gotta move it along a little bit. Uh, this prediction is the in 2025 sensitive input to a chat bot is going to be breached. Chat bots are becoming ubiquitous.

They're handling everything from customer support to personal financial advice, and every day people type sensitive information into chatbots. These might include personal identifiers, possibly financial details, almost certainly secret intellectual property. And this information is very interesting too.

Attackers, this is unintentional data leakage. We don't know about the safeguards of most of the chat bots today. We don't know if they're using end-to-end encryption.

We don't know if they're using secure storage practices. Users have a tendency to overtrust ai. People see chat chatbots is helpful as neutral, and that makes humans a little bit more likely to overshare.

The consequences of a breach like this could be devastating. And next prediction, possibly a last prediction. I predict that in 2025, there's gonna be an AI leader who's going to emerge, and they're gonna demonstrate awesome transparency and explainability.

So what does that mean? Imagine an AI system that doesn't just spit out outputs, but it actually tells you how it arrived at those outputs, providing a clear roadmap, breaking down the logic, the data which was used, the weight of each factor in the decision making process. Transparency could allow users to know what data gets fed into a system, how that data is processed, and even what potential biases were detected.

And then corrected. Explainability takes this even further. There is increasing demand for both transparency as well as explainability.

There are different stakeholders, governments, businesses, consumers, and they want AI systems that they can trust. Um, I believe that a leader will emerge in 2025 and all others will begin looking to, um, replicate similar transparency and explainability practices as this leader, uh, we're coming to the end of our time, and so I wanna share some exciting news with you. I'm writing a book, it's gonna be published in 2026, and the book is a deep dive into how artificial intelligence is reshaping cybersecurity resilience.

So there's two sides to this topic. One side is how is AI being weaponized by cyber criminals to launch increasingly sophisticated attacks? And how can AI be used as a powerful tool for defenders to build smarter, more adaptive defenses?

Um, if you wanna dive into learning more about AI and cybersecurity and you don't feel like waiting until 2026, I encourage you to go and check out my recent course available on LinkedIn Learning. If you follow these steps, you can view the course at no cost. So find me on LinkedIn, scroll down to my featured posts, select the fifth featured post 1, 2, 3, 4, 5, and that will allow you to view my 15 minute course on AI and application security at no cost.

I hope you enjoy, I hope your new year is off to a fantastic start. And thank you so much for joining me today. I.