AI in App Security: Opportunities and Risks

Security providers are constantly outdoing each other with supposedly revolutionary new products. It is difficult to distinguish between realistic promises, fear-mongering, and snake oil. Anyone who wants a better look at the state of security technology should inform themselves at a manufacturer-independent conference, such as the German edition of the Open Worldwide Application Security Project (OWASP). We spoke with Jasmin Mair and Christian Dresen, both members of the organizing team of the German OWASP Day and board members of the OWASP German Chapter. Jasmin is also the chapter lead of the OWASP Chapter Frankfurt.

There are many security conferences and events. But which topic is currently attracting particular attention in the industry—and why?

Clearly: artificial intelligence in application security. But no longer as a future vision, but as a concrete working reality. AI is no longer a hype topic for developers and security experts; the technology is already in daily use. This is precisely why we are dedicating so much space to the topic at the German OWASP Day 2025: the industry is currently grappling with what it means when AI tools are used on both sides: by defenders and by attackers. Our keynote speaker Eva Wolfangel rightly speaks of the “Code Dark Age” – a provocative term that reflects the uncertainty with which we are entering this new era of software development.

Specifically: What does the use of AI tools currently look like in the security industry?

We see two faces of AI security, which we call “Breaker” and “Builder.” On the one hand, we have “Builder AI.” This includes, for example, coding assistants that developers use daily to write code faster. On the other hand, there is “Breaker AI,” i.e., large language models used for automated attacks, sophisticated phishing campaigns, and vulnerability scans. At the German OWASP Day 2025, we will illuminate both sides: one presentation will show how the YuraScanner uses LLMs for task-driven web app scans. Another will ask the uncomfortable question of how secure the code generated by AI tools actually is. In addition, there are new phenomena such as AI agents and the Model Context Protocol (MCP), which bring their own unique security challenges.

AI tools are, of course, not only used by security experts. What further security challenges arise from the more or less visible AI hype?

The challenges go far beyond pure code. Companies often integrate AI tools faster than they understand their security implications. In his presentation “A CISO's Adventures in AI Wonderland,” CISO Holger Mack takes participants on an exploratory tour through this very jungle: How can attackers manipulate large language models? What new attack vectors does AI-driven automation open up? A concrete example from our program: the presentation “How we hacked Y Combinator companies' AI agents” shows how vulnerable some AI agents are in practice. And the secure integration of external tools into LLMs via the Model Context Protocol is anything but trivial—a real “security hot potato,” as another presentation title aptly describes it.

Attackers also immediately jump on new technology trends. What AI dangers arise from this, and how does this relate to AI regulations?

This is precisely the tension in which the industry is currently operating. On the one hand, we see how attackers use AI for increasingly sophisticated attacks—from automated vulnerability scans and targeted phishing campaigns to the manipulation of large language models. On the other hand, there is regulatory pressure, for example, through the Cyber Resilience Act, to which we are dedicating a separate workshop at the German OWASP Day. The crux: While attackers agilely exploit new technologies, companies must simultaneously meet compliance requirements and ensure practical security. This requires not theoretical treatises but concrete, practical solutions. This is exactly what we at OWASP have been delivering for over 20 years. At the German OWASP Day, we bring these different perspectives together, from technical deep dives to strategic CISO insights.

Let's get specific about the opportunities of AI: What has already become significantly better, more efficient, or more precise through the use of AI tools in security?

AI has the potential to relieve security teams, and this is already evident today. Routine tasks such as log analysis, vulnerability scans, or code reviews can be accelerated by AI-powered tools. Large language models help to recognize patterns in vast amounts of data that humans would easily overlook. This often leads to earlier detection of threats and reduced response times.

Software development also achieves higher security: Coding assistants not only suggest code faster but, when properly trained and controlled, can also point to more secure implementations. Furthermore, code analyses work well from experience: the results are not perfect, but they are generally significantly better than a development team could achieve with such reviews in the same amount of time.

The crucial point is: AI enables efficiency and precision when used correctly. It is precisely this “when” that is the focus of the German OWASP Day. We want to show how to leverage the opportunities without overlooking new risks to responsibly integrate AI into the security strategy.

One final question about your event itself: What specific content challenges are there for an explicitly manufacturer-independent conference?

Precisely because the German OWASP Day is manufacturer-independent, knowledge exchange is our priority, not marketing. However, this also means that we must select topics that bring real added value to the community and do not merely spotlight a product or tool.

This is a particular challenge in times of AI hype, as many presentations on the market today are heavily influenced by vendors. With us, the rule is: no product pitches, no sponsored talks. Instead, we rely on contributions that transparently show how security works in practice—with open tools, comprehensible methods, and real case studies.

While this independence is more complex to plan, it builds trust. Ultimately, it is the reason OWASP has been considered a credible platform for the entire security community for over 20 years—from developers to CISOs.

Jasmin and Christian, thank you very much for your answers! Interested parties can find more information on the website of the German OWASP Day 2025.

(fo)