Hackers can get top secret company information by 'confiding' with AI chatbot

According to technology experts, by using a series of sophisticated prompts, hackers can penetrate a company's 'confidential' database without being detected.

Overview of the OpenInfra & Cloud Native Day Vietnam 2025 event with the theme "Vietnam's digital breakthrough era" - Photo: CHI HIEU

Artificial Intelligence (AI) brings many values, especially now integrated by many businesses to serve end users such as company employees, customers... From here, AI-integrated chatbot systems can become a "narrow door" to help hackers attack and steal information on data storage systems.

"Currently, there are many methods of attacking AI models, the most common of which is 'content poisoning', which seems to be 'invisible' to traditional information protection systems," warned Mr. Dao Viet Hung, country director of Akamai Technologies Vietnam, at the OpenInfra & Cloud Native Day Vietnam 2025 event held on July 26 in Hanoi .

The event was organized by the Vietnam Internet Association (VIA) in coordination with Vietnam Open Infrastructure Community (VietOpenInfra), Vietnam Cloud Computing and Data Center Club (VNCDC), with the support of the Ministry of Science and Technology .

90% of businesses are using open platforms or software

In his opening speech, Mr. Vu The Binh - Vice President and General Secretary of the Vietnam Internet Association - emphasized the importance of open infrastructure and open source code in the national digital transformation process.

In the context of rapid changes in global technology, Mr. Binh said that open infrastructure continues to develop as a foundation for innovation when more than 90% of businesses are using open platforms or software in many forms.

According to a Red Hat survey (2024), 82% of technology leaders see open platforms and software as key to driving digital transformation and accelerating innovation. Platforms such as Kubernetes, OpenStack, Linux, Ceph, OpenTelemetry are increasingly playing an important role in operating AI, IoT, 5G/6G systems as well as Cloud Native.

"10 years ago, cloud could be mistaken for clouds in the sky, and open source sounded like a 'martial arts secret'. But today, we are here, connecting with the global Open Infrastructure community in more than 180 countries, together turning lines of code into the driving force for innovation," Mr. Vu The Binh shared.

AI Chatbot - "narrow door" for hackers to penetrate the system

"Currently, many businesses are strongly applying the effectiveness of open, responsive platforms that integrate systems such as AI chatbots to help improve work performance and enhance user experience.

Also from here, many methods of attack on AI models, including "content poisoning" attacks, warned Mr. Dao Viet Hung, country director of Akamai Technologies Vietnam.

He commented that Vietnamese businesses when deploying AI still do not know much about this form of attack.

Accordingly, bad actors use the method of continuously asking and re-teaching the AI ​​model false information, causing information "poisoning".

If the information on the cloud system has been "infected" and is not detected and continues to be used, it will give incorrect information or use this vulnerability to sneak into the database for the purpose of obtaining information.

Mr. Dao Viet Hung, country director of Akamai Technologies Vietnam - Photo: CHI HIEU

Mr. Dao Viet Hung cited a rare story where an auto company's online sales chatbot system was convinced to sell a new car for just $1.

The reason is that during the process of exchanging and negotiating with the chatbot, this customer found a way to "sneak" into the database and convince the chatbot to agree to the above request.

Another example is when a hacker successfully obtained all the company's confidential documents and organizational charts just by composing the right questions and sending them to the internal AI chatbot.

From that risk, Mr. Dao Viet Hung shared the AI ​​security solution so that AI can protect AI, this is a form used by many large enterprises in the world .

With this security method, the stored data integrates multiple layers of protection, making it difficult to be "infected".

The AI ​​security system can detect which information is clean and which information has potential risks of attack, from which it will have a quick and appropriate handling plan...

Back to topic

Filial piety

Source: https://tuoitre.vn/hacker-co-the-lay-thong-tin-tuyet-mat-cua-cong-ty-thong-qua-tam-su-voi-chatbot-ai-20250726155812946.htm