Vulnerabilities in LLMs: Trading security for functionality

Artificial intelligence has transformed everything from the way businesses are run to how we live our daily lives. AI automates repetitive tasks, enhances decision-making through data analysis and pattern recognition, minimizes human error, and more.
However, security researchers have recently uncovered a worrying trend: applications built with Large Language Model (LLM)-generated code and automated app-generation platforms often contain critical vulnerabilities. Why?
Simply put, these tools prioritize functionality and speed over security, resulting in exposed endpoints, insecure API calls, and applications compromised by something as simple as a curl command.
A risky business
A major drawback to LLMs and automated platforms is that they learn from internet-scraped data, where quick implementation often trumps secure design. Moreover, AI doesn’t understand business context, threat modeling, or compliance requirements.
Researchers have also demonstrated this weakness: exposed APIs in a JavaScript application enabled spam campaigns, impersonation attacks, and backend abuse - all without authentication or rate limits.
This isn’t just an academic concern. These vulnerabilities can scale quickly when organizations deploy AI-generated applications without vetting or review.
Organizations must act
It is imperative to implement threat modeling and secure coding guidelines, as no automated code generator can replace human expertise and know-how. You must also review generated code to make sure it follows these guidelines and fulfills security requirements. Lastly, continue to stress test applications to detect, and promptly remediate, vulnerabilities before cybercriminals exploit them.
LLM code that trades security for functionality is risky – can your organization afford to take that risk?
The web scanner that makes a difference
We understand the evolving risk landscape. Our web application scanner is designed to identify vulnerabilities, whether introduced by human developers or automated code generators, before they reach production. We scan all types of web applications, both self-developed and commercial applications, including for OWASP Top 10 vulnerabilities.
By integrating automated scanning into your development workflow, you gain a crucial defense layer against insecure code patterns, exposed APIs, and overlooked flaws. If your team is experimenting with these technologies, now is the time to double down on proactive security testing.
While AI-powered coding tools and app generators can accelerate development, security must never be an afterthought.

Mihail Lupan
Head of Security Research
Mihail has extensive expertise in vulnerability management and over 10 years’ experience in IT and cyber security. With a strong foundation in software development, including automation and automotive industries, he leads the Security Research team and is responsible for all vulnerability tests across the company’s suite of vulnerability scanners.