Turning Enterprise AI Chatbots into 'Fools'? Warning Issued Over LLM Cyber Threats

Emphasizing the Importance of API Security
in AI Adoption at the F5 Press Conference

"There is a new type of cybersecurity threat that could turn AI chatbots, which are supposed to serve customers, into fools."

As more and more domestic companies rush to adopt AI technologies in their businesses, there are growing calls to prepare for new cyberattacks targeting the large language model (LLM) ecosystem. F5, a company specializing in application security, held a press conference at the Park Hyatt Seoul in Gangnam-gu, Seoul, on the 23rd to discuss security measures for LLMs in the AI era.

Mohan Bellu, CTO of F5 Asia Pacific, China, and Japan, is explaining new cyber threats in the AI era at a press conference on the 23rd. Provided by F5

Founded in 1996, F5 operates in over 170 countries, and its Korean branch is marking its 25th anniversary this year. The company has more than 23,000 customers worldwide, and 85% of the Fortune 50 companies use F5's security technologies.

Mohan Bellu, CTO of F5 Asia Pacific, China, and Japan, explained, "As AI technology rapidly advances, new cyber threats are emerging, including hallucination issues, personal information theft, prompt injection, and the malicious use of LLMs."

James Lee, Senior Solutions Architect for F5 Asia Pacific, China, and Japan, said, "When banks or hospitals introduce AI chatbots to serve customers, they are assigned specific roles. However, hackers may assign new tasks to these chatbots, turn them into fools, or exploit them in other serious ways."

CTO Mohan Bellu also cited a case last year in which an AI chatbot operating on the Air Canada website provided inaccurate airline ticket information to customers, resulting in a civil dispute. This shows that AI chatbots, which are introduced to improve operational efficiency, can become a risk for companies.

James Lee, Senior Solutions Architect for F5 Asia Pacific, China, and Japan, stated at a press briefing on the 23rd that AI companies based on LLM will face new security threats.

According to F5's "2025 Application Strategy Report," 96% of companies are deploying AI models. It is predicted that within three years, 80% of all applications will utilize AI. However, F5 noted that most companies are not fully prepared to effectively handle the vast data, complex traffic patterns, and security threats inherent in AI applications.

Architect James Lee stated, "Without API security, it is impossible to secure AI models. When DeepSeek was released, there were reports that it had illegally collected data from OpenAI, and this is also related to API security, which governs the calling and exchange of data."

He emphasized, "F5 solutions identify vulnerabilities, strengthen security, and detect and resolve threats before attackers do, thereby effectively protecting both existing and new AI-based applications."

Meanwhile, F5 will host the "F5 Solution Day 2025" event on the 24th at the Sofitel Ambassador Seoul, inviting about 300 domestic and international IT experts. The company plans to introduce ways to enhance corporate competitiveness through partnerships in an AI-driven environment.