Expert: AI makes traditional security methods reveal limitations

(Dan Tri) - AI is helping hackers shorten the time to search for vulnerabilities and network attacks, posing new challenges for businesses as more and more data is uploaded to the cloud computing environment.

Talking to Dan Tri reporter, Mr. Phan Hong Tam, Cloud Technology Director, FPT Smart Cloud Company (FPT Group) said that the emergence of AI not only changes the way of cyber attacks but also forces businesses to reconsider their ability to protect their systems.

According to experts, when hackers can take advantage of AI to automatically search for vulnerabilities and carry out attacks at high speed, traditional defense methods are revealing many limitations.

Notably, the development of AI agents also poses new risks in access rights management and enterprise data protection.

Mr. Phan Hong Tam, Director of Cloud Technology, FPT Smart Cloud Company (FPT Group), warned of new cybersecurity risks as hackers increasingly take advantage of AI to attack (Photo: FPT Cloud).

AI makes cyber attacks more dangerous

When AI is used by hackers to automatically search for vulnerabilities, create malicious code and carry out phishing campaigns, how has the speed and scale of cyber attacks changed compared to before, sir?

- In my opinion, the biggest change lies not in the emergence of a new type or method of attack, but in the speed and scale that that threat brings.

Previously, to detect targets, try many attack variations or build a phishing campaign, the implementer needed a certain amount of time and qualifications.

However, now AI is significantly shortening those steps and allowing these activities to be performed continuously at a much lower cost.

Recent international reports reflect this quite clearly. CrowdStrike's Global Threat Report 2026 said that in 2025, the average time for an attacker to go from the first point of entry to other systems is only 29 minutes, the fastest case is 27 seconds; The activity of attack groups using AI increased by 89% compared to the previous year.

The more AI increases the attack speed, the more businesses must increase their defense speed accordingly.

If the attacker is already operating at machine speed, but the defender is still waiting for humans to read each warning and then handle each case, the risk of slow detection or missing unusual signs will increase, creating conditions for the attacker to expand the scope of intrusion, escalate privileges or access more important systems and data.

According to Mr. Tam, artificial intelligence (AI) is helping hackers shorten the time to search for vulnerabilities, expand the scale and speed up network attacks (Photo: FPT Cloud).

From actual deployment and monitoring in Vietnam, what are the most common forms of attacks that the unit records? How is AI making these attack methods more dangerous?

- From actual deployment and monitoring in Vietnam, we recognize common forms such as DDoS, supply chain attack, exploiting third-party software vulnerabilities and phishing; Besides, misconfiguration is still a common weakness.

AI support is making these types of attacks more sophisticated and faster.

For example, phishing emails and messages in Vietnamese can now be created with a very natural style, with few errors and targeted at the right audience, making it more difficult for users to recognize.

With software vulnerabilities, the time between when a CVE is announced and when it is automatically scanned and widely exploited is becoming increasingly shorter.

Malware variants can also change more rapidly, reducing the effectiveness of defenses based only on known identification patterns.

AI therefore does not necessarily cause a sudden increase in the number of attacks in a short period of time, but is helping attackers shorten preparation time, expand scale and increase the ability to hide their behavior.

In cloud computing infrastructure, what are the security weaknesses that Vietnamese businesses are most likely to miss?

- In digital infrastructure, there is no "absolutely secure" layer. From identities, APIs, data, workloads to management systems, any layer can become a target if there is misconfiguration, weak control mechanisms or lack of supervision.

Therefore, businesses need to assess risks in layers, instead of just focusing on one area that is considered most vulnerable to attack.

Among the above items, the management system is the area that Vietnamese businesses are most likely to overlook, because this is where many other components are controlled such as identification, authorization, workload and data.

If administration policies, configuration or decentralization are not tight, an exploited point can quickly expand the scope of influence to many other systems.

Mr. Tam said, AI agents have the ability to perform tasks on their own, access data and interact with the system, but also pose a potential security risk if given too broad permissions (Photo: FPT Cloud).

As AI agents become increasingly capable of performing tasks on their own, accessing data and interacting with systems, could this technology become a new security weakness? How do businesses need to control these risks?

- AI agents can completely become a new weakness if given too broad authority or lack appropriate control points.

Unlike AI that mainly creates content or makes recommendations, AI agents can make plans, call APIs, access data and perform tasks on the system.

As the level of autonomy increases, the risk is no longer just a “wrong answer”, but can turn into incorrect actions, out-of-range access or impact on other systems.

The IBM 2025 report shows that 13% of organizations have recorded a breach incident related to an AI model or application; of these, 97% do not have AI-specific access control mechanisms.

That shows that the big risk often lies not in the model itself, but in the way AI is connected to data, tools and systems but is not properly governed.

Enterprises need to manage AI agents as an entity with its own identity and access rights: grant only the necessary permissions, limit the scope and time of use, and fully record actions so that they can be checked and traced.

According to you, in the next 3-5 years, how should Vietnamese businesses build a cloud security strategy to proactively respond to AI-supported attacks?

- If building a Cloud security strategy for Vietnamese businesses in the next 3-5 years, I think it is necessary to change the approach from investing in more security tools to building integrated, proactive and automated defense capabilities.

I propose a strategy in three directions:

Secure by Design: Security is designed into the architecture, rather than added after deployment.

AI-Native Security: Use AI to detect, analyze, predict and support handling security risks.

Self-Driving Security: Moving towards the ability for the system to self-identify risks, make its own decisions within policy limits, and self-enforce allowed defense measures.

The goal is not just to apply AI to the SOC, but to redesign the entire defense lifecycle: detect earlier, react faster, limit the scope of impact and continuously learn from incidents to improve protection capabilities.

Thank you for your sharing!