Threat modeling with GenAI

Making Threat Modeling Scalable with GenAI

Recognize risks earlier in the design

Threat Modeling with GenAI: Scaling Security by Design
21.04.2026
Security
Artificial Intelligence
Application Modernization
Application Development

Releases are getting faster, security reviews often struggle to keep up. Threat modeling is effective, but is difficult to scale manually. This article shows how GenAI can help anchor threat modeling earlier in the design process and relieve the burden on security experts in a targeted manner.

Threat Modeling as the Basis for Security by Design

Threat modeling is a structured method for identifying at an early stage what risks a system faces, how attacks might unfold, and what protective measures should be implemented as a result.

 

This approach fosters a shared understanding of what needs to be protected, which components are security-relevant, and how a system can be attacked. This is precisely where its value for Security by Design lies: risks become apparent not only at the end of a project, but as soon as architecture, data flows, and trust boundaries are established. This allows security measures to be specifically derived, prioritized, and integrated into the development process.

 

In practice, threat modeling can be structured well using the four key questions according to Adam Shostack:

  • What are we working on?
  • What can go wrong?
  • What do we do about it?
  • Have we done a good job?

 

This framework is as simple as it is effective. It helps teams not only discuss individual vulnerabilities but also systematically incorporate security risks into architectural and design decisions (source: The Four-Question Framework).

Why Threat Modeling Often Becomes a Bottleneck

It is precisely because threat modeling is so effective that its lack of scalability is particularly significant in many organizations. Traditionally, it is an expert-driven discipline: security architects, security engineers, and delivery teams work together to analyze system designs, data flows, assets, and trust boundaries. This makes technical sense, but is difficult to implement consistently in dynamic environments with short release cycles, cloud services, and many teams involved.

 

The consequences are well known: Threat modeling happens too late, too superficially, only selectively, or exclusively for particularly critical systems. This is where the real problem arises. Not because the benefits of threat modeling are in question, but because specialized expertise is scarce and manual analyses only scale to a limited extent. Risks identified early in the life cycle are generally easier and more cost-effective to address than late corrections shortly before go-live.

How GenAI Can Support Threat Modeling

GenAI does not change the technical logic of threat modeling. However, it can help to apply it more consistently and with less manual effort.

 

A meaningful application begins where architecture artifacts, system descriptions, or initial design assumptions are already available. GenAI can use these to structure the security-relevant building blocks: Assets, data flows, interfaces, and trust boundaries. This does not replace a technical review, but it creates a reliable starting point for further analysis.

 

Based on this, GenAI can systematically derive threats. The use of established models such as STRIDE to structure threats along individual components or data flows is particularly helpful here. STRIDE helps to classify potential risks in a comprehensible manner and derive suitable countermeasures.

 

Another advantage lies in working with reference sources. If GenAI is linked to curated knowledge sources, identified threats can be compared with known vulnerabilities, attack techniques, or established security patterns. CVE, NVD, or MITRE ATT&CK are suitable for this type of classification. This turns a purely generated analysis into a more comprehensible and referenceable work status.

 

It is important to set the right expectations: GenAI does not automatically deliver "correct" threat models. It helps to work more systematically, recognize blind spots earlier, and increase the quality of the discussion. The assessment of business risks, residual risks, and priorities remains a professional and organizational decision.

Comparison: Threat Modeling Classic vs. With GenAI

The direct comparison between classic threat modeling and an approach supported by GenAI shows exactly where the difference lies:

 

AspectClassic threat modelingThreat modeling with GenAI
ExecutionHeavily manual and workshop-basedMethodically supported, partially pre-structured
Dependence on expertsHighStill high, but more targeted
Entry barrierRelatively highLower, because structure and preliminary work are supported
SpeedTime-consuming, especially for complex systemsFaster preparation and more structured analysis
ConsistencyDepending on experience, team and available time budgetHigher consistency in structure and documentation
Identification of threatsStrongly dependent on individual perspectiveSystematic matching of patterns and variants
How to deal with blind spotsRisk of tunnel vision or confirmation biasAdditional perspective on assumptions and gaps
Traceabilityoften heavily dependent on documentation qualityBetter structurable and enrichable with references
ScalabilityLimited across many projectsBetter scalability, especially in early phases
Role of the security expertCentral operational and technical roleCentral technical role, more focused on evaluation and prioritization

 

The comparison shows: GenAI does not make threat modeling superfluous, but more structured and more scalable. Security experts remain crucial - but their focus is shifting more towards evaluation, prioritization and well-founded risk decisions.

From Threat Model to Design Decision

The true value of threat modeling lies not in the list of identified risks, but in translating those risks into design decisions. The threat model gives rise to requirements for architecture, interfaces, permissions, segmentation, the protection of data flows, and the handling of failure scenarios.

 

This is precisely where threat modeling becomes part of the secure-by-design process: risks are not only identified but also translated into concrete architectural and design decisions.


 

This is where threat modeling and secure-by-design converge directly. Design principles such as least privilege, defense in depth, or secure defaults only deliver their full value when applied within the specific system context. This is exactly where GenAI can help: by relating relevant principles to the respective design, making gaps visible, and discussing mitigation measures earlier in the design phase. Secure by Design and Secure by Default pursue precisely this goal: not adding security as an afterthought, but embedding it in the product and the development process from the very beginning.

 

This is particularly valuable for teams because architectural deviations become visible earlier. This often allows avoiding later corrections, which are significantly more expensive and prone to conflict when done under time pressure.

Reviews with Methodological Feedback

Threat modeling doesn’t end with the initial analysis. The results must be reviewed, updated, and aligned with design and implementation decisions. This is precisely where the fourth guiding question comes into its own: Have we done a good job?

 

Reviews often reveal a classic problem: Teams naturally evaluate their own assumptions differently than an external observer would. GenAI can be useful here as an additional perspective. It can re-evaluate mitigation chains against the threat model, highlight implicit assumptions, and make inconsistencies between architecture, threat assumptions, and countermeasures more clearly visible. This does not replace an experienced reviewer, but it improves the transparency, comparability, and reproducibility of the discussion.

Context Data as a Prerequisite for Ai-Supported Threat Modeling

The decisive factor for getting started with AI-supported threat modeling is not the formulation of tasks, but the availability and usability of the relevant context data. If this is available, a few clearly formulated tasks are often sufficient:

  • "Identify the trust boundaries in this architecture or data flow diagram."
  • "Apply STRIDE to this component and prioritize the most important risks."
  • "What mitigation measures are missing to ensure that this design adequately addresses the identified threats?"

 

The results must then be professionally checked, supplemented, and evaluated in the context of your own architecture.

Conclusion

Threat modeling is an effective tool for reliably implementing security by design in day-to-day operations. The problem isn’t the method itself, but its limited scalability in complex delivery environments. This is precisely where GenAI comes into play: it structures the threat modeling process, lowers the barrier to entry, makes analyses more consistent, and relieves security experts of time-consuming routine tasks.

 

The key point, however, remains unchanged: GenAI does not replace security experts. It supports them. Effective threat models emerge where methodological structure, reliable references, and professional judgment come together. Those who use GenAI in this way not only strengthen threat modeling operationally but also embed security earlier and more effectively into the design process.

Frequently Asked Questions About Threat Modeling with GenAI

  • GenAI can prepare threat modeling in a more structured, consistent and faster way. It helps to prepare architectural information, systematically derive threats and compare results with known reference sources. The greatest benefit lies not in automation for its own sake, but in making the method more scalable in everyday use.

  • No. Threat modeling remains a technical discipline. GenAI can do the groundwork, make patterns visible and structure analyses. The assessment of risks, the prioritization of measures and the decision on residual risks remain a human responsibility.

  • GenAI can help not only to identify risks, but also to translate them into design decisions at an earlier stage. It makes security-relevant correlations in the architecture, in data flows and at the trust boundaries more visible and supports teams in taking suitable countermeasures into account as early as the design stage. This links threat modeling more closely with secure design processes.

  • On three points: sufficient quality of the input data, clear guidelines for data protection and compliance as well as proper embedding in existing architecture and review processes. The clearer the scope, reference sources and responsibilities are defined, the more reliable the results will be.

  • GenAI can prepare and structure analyses, but cannot replace a professional risk decision. Results must be checked, categorized and evaluated in the context of the specific architecture. Human judgement remains crucial, especially when it comes to protection requirements, residual risks and prioritization.

Further Information on Security, GenAI, and Application Development

Application Development

Tailored application development allows you to create more efficient business processes, seamlessly integrate new applications and optimize scalability by leveraging API Economy and AI Infused Apps.

Application Modernization

We transform your specialist applications. Modern, modular, secure.

Generative AI

Generative AI such as ChatGPT and Midjourney can be used to completely rethink existing value chains. Arvato Systems supports you in this.

How the Secure Development Lifecycle Protects Against Threats

Discover how to integrate security requirements early and build a robust IT security culture. Learn the fundamental principles of the Secure Development Lifecycle and effectively protect your applications from security vulnerabilities.

Secure by Design: Principles for a Secure Architecture

The article shows why security by design is a central architectural principle and how security is anchored in the design of modern IT systems.

Written by

Jungnickel_Erik
Erik Jungnickel
Expert for Application Development

Erik Jungnickel is a software developer with over 15 years of practical experience. His focus is on the front-end development of modern web applications and the use of AI technologies with Microsoft Azure. He designs innovative and future-proof solutions by combining technical depth, cloud know-how, and practical experience.

Learn more about this author