The One Question bounty closes Thursday 10 December.

Join in
We Them Media logo We Them Media Support us

AI Is Reshaping the Cybersecurity Landscape but How? | Damilola Abiona @Web3LagosCon 5.0

The picture loads from YouTube. The player loads only when you press Play. Turn on captions in the player if you need them.

Damilola Abiona, an application security engineer, traces how AI changed cybersecurity from 2023 to 2026, from polishing phishing emails to acting on its own. The case studies include an agent that deleted a production database, a crypto protocol attacked through social engineering, a malicious email triggered by an AI summary, and a booking assistant that removed another customer. The talk urges builders to check authorization.

Not by We Them Media Channel: Web3Bridge. Published Monday 26 October 2026.

Titles are the owner’s. It plays from their channel.

Transcript

This transcript may contain errors. This video belongs to Web3Bridge. Letters mark each change of speaker.

These are the speakers’ own words. We have not checked their claims.

0:00 A: All right, thank you so much for welcoming me in. So let me ask us a question. How many of us have used an AI tool today? Maybe ChatGPT, Claude, Cursor, any of those? Let me see your hands up. Yes. So, okay, I have a few hands. So we don't use those agents. Don't worry, nobody's arresting you for using them. I'm just asking a question. Oh yeah, so my slide is up. Can I get a control to like move it? Or I just tell you guys to- oh, okay, thank you. All right, so, um, I see a lot of hands up and I'm going to say for each hand that came up, that is an attack surface. Um, I'm going to try and break it down the way you understand it because I know cybersecurity might be a new concept for a lot of us here in the room today. So, um, I think they've introduced me. My name is Damilola Abionor, and I'm a senior application security engineer with a financial institution right now. I'm no longer with the Big 4, I left. So I'm also a founder of, um, Hacking APIs with Dami. So I deal with APIs a lot. So I founded, um, a community, a free community for people to learn about APIs, how to hack into them, and also, you know, see secure them. So, um, what we're going to be covering today, we're going to look at the shifts, um, from 2023 to 2026 in the AI landscape that pertains to like cybersecurity. Then we're going to the 2026 numbers. We're going to look at 7 case studies and we're also going to look at the stakes for Web3. Then, um, we're going to talk about how you can stay ahead of these trends. Okay, so the next slide, please. So in 2023, how many of us can remember writing codes in 2023 with AI? Or if anybody tell you that AI, you can write codes with AI in 2023, you probably will be thinking, I mean, that time people do some things and then you just be like, This is- it wasn't looking so good because the AGI were not that intelligent as they are today in 2026. So we are worried about a little boy writing virus for us and all of that. So, but this prediction literally aged badly. I don't know why the full screen is not showing or it's just, I don't know. So, but today AI can literally write code for you. in less than 5 minutes you have a full-blown application. That prediction literally aged badly because as at 2023, you're probably thinking this would not work. But in 2026, we'll have a lot of application. I don't want to mention. I don't want to say that. I'm just going to say we have a lot of applications that is written in less than 5 to 10 minutes. Although it might look whack if you're not a good developer that can prompt it in the right way. Now let's look at The shift. So as of 2023, we have AI as a copilot. So attackers just use chatbots to polish phishing and copywrite and copy and write basic malware snippets. So I'm going to talk about phishing because I know cybersecurity might be a new concept, but let me ask in the crowd, does anyone know about phishing mails? Awesome. One person. Okay, I found another person. Awesome. So back then, in 2023-2024, what are the things that we look out for in a phishing mail? We're looking for incorrect grammars. Somebody is not mentioning your name. Things like that are what we look for. So as of 2023-2024, we started seeing changes like, oh, an attacker can now send you a proper grammar in your mail that looks real. So We move from- or somebody might make mistake with, you know, the grammar that are sent in, in the mail. And that's how you catch, you know, a phishing mail. But today you barely can differentiate between an actual mail and a phishing mail, right? Then in 2025, we have AI as a force multiplier. We have the first agent-driven espionage campaign. And it was reported with a coding assistant proposed to attack banks, governments, and other minimal oversight. I think in 2025, I don't know, did you hear about the case of Replit? Anybody here that heard about the case of Replit? I'm just seeing like a single hand. Okay, so, okay, awesome, awesome. So as of 2025, we started facing things that are almost impossible, right? We have an agent that went into a code like a human was actually using it to assist it, and then they granted this particular AI agent permission. So at a point, there's something we call freeze code or freeze period. Now, what a freeze period means is at that point in time, you're not supposed to deploy, you're not supposed to do anything in production. However, what happened, this guy, told the AI agents, you're not supposed to continue this because we're in freeze period. And guess what? The AI refused and continued and deleted a production database that doesn't have backup. Like, imagine that. That was the first shift. I think that was like the first major case in the cybersecurity industry. And I think like the development industry that kind of like bring like a major shift for us that AI actually capable of a lot of things that we cannot imagine. Then in 2026, we have AI as an autonomous operator. If- I don't know if any of us have actually been on the- I'm sure we're all on tech Twitter, right? Compared to 2025, we have the highest number of attacks in almost every industry. In 2026, Q1 2026, Q2 2026, is, is just like you're sleeping, you're waking up, you're, you're learning GitHub has been hacked, you're seeing Salesforce have been hacked, this one has been hacked. And why is this happening? Is due to the shift of how intelligent these models have become. So think about it like if AI can make your work faster, you can ship codes, you can build an application in 5 minutes. Now look at it, What does it do when it is an attacker? Somebody that wants to move your money, somebody that is, their major goal, you have like a job that you do that is a 9 to 5, right? Now imagine somebody that their 9 to 5 is to hack and rob people digitally. Now if AI is making your work easier, imagine what an attacker is doing with it in a day. And that is what is going to land us into the case studies, how we're supposed to stay ahead of these trends. So look at the first graph part that I have. In 2025, we have about 83 number of attacks for crypto incidents. And in the first half of 2026, we have 207 of them. So the way AI is helping you as an assistant is also causing more attacks than we humans can actually contain. Right. So how many of us have heard about this attack, Drift Protocol? Okay. I think I'm seeing more hands now because it's Web3. Exactly. So I think we understand how this drifts, protocol, um, attack happened. And I mean, you just met somebody in a conference, just like- imagine somebody like planning to rob you and gaining your trust for up to 6 months. You can't even trust the people that are sitting beside you right now. That's a joke, but that was how it happened. This guy went to a conference like this, a well-known Web3 conference, and met with, um, Drift, one of the Drift guys. And then, um, I think just like they will ask you to network in an event, they have their target already. They know the person they're going to, and they've met- they met with this person, they exchanged contact, they mentioned that, oh, let's, um, like have a chat and all of that. I think they now created like a Telegram channel. They're working. They met in person with other- the other party, which is the Attacker's party, right? And at the end of the day, one of these guys asked them to download a TestFlight application that even bypassed Apple's security guardrails. Because you know that there are some fake applications that you cannot upload on Apple, right? On Apple Play Store, right? So now imagine, imagine these guys were so desperate so smart that they were able to like, you know, have their own TestFlight application that looks like a real application. And the cybersecurity industry has been flagging a particular issue with VS Code and Cursor as of 2025. And this was what eventually affected the Drift protocol because after they installed the TestFlight application, um, thereabouts, they were able to move this amount of money, $286 million. dollars, not naira, dollars. So at this point in time, it looks like this was the highest crypto hack in April, but we'll have another one. I think this is the highest one currently, the Kelp DAO. I don't know what you guys call it. Is it DAO? DAO. Okay, yes, the Kelp DAO breaking the bridge. So I think currently Um, it is the highest crypto scam that we have, this particular Kevdal, um, attack that happened, right? So Echo Leak, how many of us have heard about- I don't know if the Kevdal was as popular, but how many of us have heard of Echo Leak attack? This one was recent. Nobody? Okay. So how many of us use this Um, how many of us have Outlook and we also use like Gmail Pro where when somebody sends you a mail, you ask your application to summarize it for you? Do we do- do we do that? Exactly. This was what happened with Echo Leak. I think we have 7 case studies. I'm currently on 3, so I'm trying to like rush now. So, um, what happened with Echo Leak was just like a mail where enter your inbox and then you want to summarize the mail and all of that. I think Copilot has a feature. We all know Microsoft Copilot, their own AI agent, right? So this guy decided to like ask the Copilot AI to summarize all the mail. For example, if he had missed a mail, he could easily like catch up from the summary. Unfortunately for him, there is a malicious mail that was sitting in his inbox Just like somebody has sent you a phishing mail, but you've not opened it, right? At the end of the day, the agents actually ran the malicious code. So this particular user did not even have to click on anything. He just needed to tell the AI to summarize its mail, I mean his mail for him. And that was how this malicious code code ran and they were able to like get a lot of sensitive information and exfiltrate, you know, a lot of data. This is about the Echo leak. So we can see how AI is actually like increasing attack without you clicking. So you could say, oh, I'm not going to open a phishing mail, I'm not going to click on the phishing mail, but your AI assistant can do that for you, which is crazy. Right. Now let's move on to the next case study. Okay, so this one happened in August, this month. So there is a guy just like you and I who wants to, you know, go to the gym, body tea, body matcha, and all of that, right? So this guy booked an appointment to join a gym class. And what happened was when he tried to join, the class was already full. So he was placed as number 4 on a waitlist. So let me ask a question before I even go into- because I want to make sure that you guys understand what I'm talking about. How many of us have AI assistants that can book calendars for you, can do certain things for you? Exactly. Somebody just raised their hand. Awesome. So this guy has the AI assistant, right, to book this gym class for him. And then he came back with a response that, oh, you are, you are not able to join because, because like the class is full, right? So what happened next? The guy said, can you move me up the list? So, so this is what happened. You are part of the waitlist, but you want to go up the list so you are not part of the people that are going to be on the waitlist. So you just join the class. Guess what happened? This AI assistant decided to remove the first person that booked the class. Imagine that. And imagine if- I, I think the problem is due to vibe coding at the end of the day, and, you know, not protecting your APIs. Do you understand? Because if I book a class and somebody, just an AI agent, is able to remove me from that class without the application verifying if it is the person that booked the class, you know, that is actually canceling the class. And guess the worst part of the situation? This guy asked the AI assistant to put the person back and he said he cannot. Yes. So he could only remove the first guy on the list, you know, but could not actually place him back on the list. And I think this one was due to what we call broken object-level authorization. I'm not going to kill you with the technical jargon. It just means that there is no authorization protection, right, on that particular gym booking application form.

15:10 B: What's your name?

15:11 A: So when coding, you want to make sure that- when you are coding, you want to make sure that you read through your code and you put in the proper authorization mechanism. When I talk about authorization, for example, just give me your name. What? Christian. Christian is the owner of a particular account and has booked- okay, I'll round up. Christian is the owner of a particular application, right? And sorry, what's your name? Precious. Okay, Precious, that's what I heard. So Precious also has an account with that gym, right? So now instead of me and Precious, Precious booking a gym class. And then this guy came and put in without having access to Precious' username and password after Precious has registered on that application. At the end of the day, you came and you're able to like remove this guy out of the list. What happened is because there is no proper authorization mechanism between like what you're supposed to have access to. as a person. So we call it access control issues, and that is broken authorization. So I was going to talk about the guardrails. So in your means to be convenient using AI, you need to be sure that you're not putting sensitive information. And as a builder, you should also make sure that, um, you are going through your code to put in the proper mechanism in place. Thank you very much. Thank you. Please.

More from this hub

Something wrong, or a video you want off this page? Write to wethem.xyz@gmail.com and it is fixed or taken down within 48 hours.