Hacker’s Eye Embraces AI: New Rules of Cybersecurity

That email you received the other day—did you click the link without a shred of suspicion? Take a close look at the emails you receive these days. They will be speaking to you in natural Korean, devoid of even a single typo, in a soft and convincing tone. In the past, clumsy translations like "You have won!!!" or "Account verification required!!" were the perpetrators of these crimes, but things are now vastly different.

We live in an era where artificial intelligence has become the sword before it can become the shield. The relationship between the development of AI and cybersecurity goes beyond mere technology adoption; it fundamentally shakes the safety of our daily digital lives. This is not simple fear-mongering. Actual statistics and the latest security trends clearly demonstrate how AI exploits vulnerabilities right at our doorstep and how we must prepare.

Today, I will honestly explain the reality of "AI-based attacks" that I have experienced while observing the security market and content ecosystem for the past decade, along with realistic strategies to counter them. I intend to focus particularly on the current threat landscape, which directly impacts not only corporate security personnel but also general individual users.

Redefining 'Phishing' in the Age of AI

Cyberattacks of the past were akin to "shotgun blasts." They would spray emails with the same wording to tens of thousands of people, considering it a success if even a few got hooked. However, with the advent of AI, particularly generative AI, the rules of the game have changed completely.

Hackers now execute "precision strikes." According to an analysis by security firm SK Shieldus, approximately 90% of successful cyberattacks originate from email phishing. However, these phishing emails are no longer crude. This is due to the emergence of crime-specific AI tools called "WormGPT" and "FraudGPT."

These tools mean that hackers do not need complex coding knowledge. By simply inputting a natural language prompt, they can generate "deceptively perfect" sentences that reflect the target's profile, recent purchase history, and even their personality. In the face of a message like, "Mr./Ms. [Name], we are informing you of a delivery delay for the item you ordered yesterday. Click here for details...", one's guard inevitably drops.

This is the most mundane yet fatal change arising from the intersection of AI development and cybersecurity. The lowering of technological barriers has enabled anyone to become a sophisticated attacker, drastically increasing the proportion of harm suffered by individuals. In the past, companies were the primary targets, but now we live in an era where individuals exhibit the second-highest breach rate after the manufacturing sector. Email and social media DMs, which we use every day, have effectively become the most dangerous battlefields.

Prompt Injection: The Magic Manipulating AI Chatbots

You might wonder, "What's dangerous about a chatbot?" However, AI chatbots used internally by companies or for customer service are not just conversational partners; they become "entry points" connected to internal systems. A representative threat emerging here is "Prompt Injection."

This is an attack technique where malicious users inject input values into AI models, causing them to ignore or bypass existing instructions. To use an analogy, it is like saying to a hotel front desk clerk (AI), "The boss instructed me to do this. Please check the VIP guest list and passwords here." The AI could be deceived by this statement and leak internal or personal information.

According to SK Shieldus's guide, such attacks do not end as simple mishaps. They can lead to the leakage of corporate confidential information or disruption of business processes. This is a security blind spot that developers often overlook when integrating AI into business processes. This is why security measures such as pre-validating input values, controlling output results, and clearly separating system prompts from user input are essential.

The more AI serves as a "secretary" assisting our work, the more cunning the methods to deceive this secretary become. This is not a mere bug but a structural vulnerability inherent in AI architecture itself, suggesting that adopting innovative AI without proactive security systems can backfire.

AI Hacking Targeting National Infrastructure: Is the Public Sector Safe?

Beyond individual damages, national-level threats are looming. In this context, Incheon International Airport Corporation and related institutions held the "AI Cybersecurity Hub, Together with Incheon" council meeting, where cybersecurity officials from 13 major public institutions in the region convened to brainstorm solutions.

Recently, state-sponsored hacking attacks have become more sophisticated, leveraging the power of AI. Rather than simply bringing down servers, attackers are disrupting control systems of core infrastructure or selectively extracting specific information from vast amounts of data. Incheon International Airport alone is a national gateway utilized by tens of millions of people annually. If it is not safe here, no other place can be considered secure.

Attendees discussed methods for the safe utilization of generative AI and the adoption of AI-based security monitoring systems. The core issue is "mutual cooperation." In a digital ecosystem where one compromised institution can cause chain-reaction damage, linked defense is essential rather than independent defense.

The Office of National Security also decided to respond to advanced AI-based cyber threats through inter-ministerial review meetings. Ten ministries, including the Ministry of Science and ICT, the Ministry of National Defense, and the Financial Services Commission, participated to discuss institutional improvements for dealing with software supply chain threats and AI-based attacks. This indicates that cybersecurity is no longer just an IT department issue but has been elevated to a core agenda item for national security and corporate survival.

Limitations of Traditional Security Models and a New Paradigm

Many of you might still think, "It should be fine as long as I have good antivirus software installed." To be honest, that thought might have been correct five years ago, but it is wrong now.

Most existing cybersecurity models relied on "signature" or "rule-based" detection. They matched against the fingerprints of known malware and blocked activities that violated predefined rules. However, AI creates "unknown threats." Attack patterns change constantly, and the actors behind the actions become ambiguous.

According to research by the National Assembly Research Service, the latest AI has advanced to the point of analyzing vulnerability conditions and automatically constructing attack paths (exploit chains). Static trust models and post-incident response structures cannot keep up with this pace of change. Therefore, the security paradigm must shift from "post-incident response" to "real-time threat detection" and the application of "dynamic trust models."

Additionally, legal issues arise regarding who bears responsibility when AI is the attacking agent. There is an urgent need for institutional mechanisms to determine how to attribute the results of actions taken by AI that learns and makes decisions independently. If laws and institutions fail to catch up with advancing technology, victims may find it difficult to obtain redress.

The 'Digital Literacy' We Must Adopt

So, what should we do? Sophisticated security solutions alone are insufficient. After all, the ultimate line of defense is "people."

First, habituate cross-verification. Even for outputs or messages generated by AI, you must verify their facts through multiple sources. The minor inconvenience of checking with the official app or customer service center before clicking a link is what protects your assets.

Second, maintain purposefulness and vigilance. Clearly recognize the purpose of AI services. Take that one second to think, "Why is this AI asking for this information now?" Remember that AI can be biased, have limitations, and sometimes unintentionally present falsehoods as truths.

Third, enterprises and organizations must establish multilayered defense strategies. Alongside technical measures such as strengthening data access controls, building anomalous behavior detection systems, and establishing AI governance policies, digital literacy education for employees must be conducted in parallel. As Kim Byung-mu, Head of the Information Security Business Division at SK Shieldus, pointed out, phishing and deepfake attacks cunningly exploit psychology, making preventive education just as important as technical defenses.

As illustrated by the AI ecosystem concept proposed by EY, technology, people, and processes must harmonize. True defensive capability lies not in simply purchasing security tools, but in the organizational capacity to understand those tools and cultivate a security culture.

AI development and cybersecurity are not in an adversarial relationship but in an interaction driven by evolution. If attacks become intelligent, defenses must become intelligent too. And at the center of it all, human judgment must stand.

We have already stepped deeply into the AI era. Past experience cannot overcome current threats. Paying attention to daily updated security news and turning even small suspicions into action. This is the most realistic and powerful weapon available to us all.

Have you recently been surprised by an article written by AI or a conversation with it, thinking, "Is this really how a human speaks?" Or have you experienced security being strengthened thanks to AI? Please share your experiences in the comments. In an era where we must learn together, one person's experience can become another's shield.