Skip to content
    Flow-alerta.cab

    Adversarial Robustness

    AI fundamentals & modelsRobustness against adversarial examples
    Editorially reviewed on: 3 sources
    Your reading view

    The short answer · 13 words

    In short

    Adversarial robustness describes the resistance of an ML system to targeted manipulated inputs.

    Flow-alerta.cab (2026): Adversarial Robustness. https://flow-alerta.cab/en/glossar/adversarial-robustness. Editorially reviewed on: 09 October 2026.

    How much detail would you like?

    Practical

    Important for AI security in systems that process external or potentially manipulated inputs.

    Possible example: An image classifier is tested with FGSM or PGD-like attacks and additionally evaluated on unchanged data.

    Potential applications: Red teaming and robustness tests for classification, document or image recognition systems before productive use.

    At a glance

    Core
    A robust model should also react as correctly or reliably as possible when attacking inputs change in a targeted manner so that misclassifications arise. Robustness is always evaluated relative to a defined threat model.
    Use
    Important for AI security in systems that process external or potentially manipulated inputs.
    Example
    Possible example: An image classifier is tested with FGSM or PGD-like attacks and additionally evaluated on unchanged data.
    Limits
    Interpreting robustness against a single attack as general security; ignoring unrealistic threat models; ignoring clean accuracy; not considering adaptive attacks.

    In context

    Adversarial Attacks

    Adversarial Attacks are targeted attacks on ML systems in which input or other components are manipulated so that the model produces unwanted or incorrect results.

    AI Safety

    Design and deploy AI systems to prevent or limit damage.

    Questions about this term

    Is robustness absolutely measurable?

    No, it depends on the threat model and attack.

    Does adversarial training make a model safe?

    It can increase robustness but does not guarantee general safety.

    All details and background

    A robust model should also react as correctly or reliably as possible when attacking inputs change in a targeted manner so that misclassifications arise. Robustness is always evaluated relative to a defined threat model.

    ELI5 – Explained simply

    AI should not be tricked by small, deliberately chosen changes.

    Why it matters in practice

    Important for AI security in systems that process external or potentially manipulated inputs.

    Example

    Possible example: An image classifier is tested with FGSM or PGD-like attacks and additionally evaluated on unchanged data.

    Pitfalls and limits

    • Interpreting robustness against a single attack as general security; ignoring unrealistic threat models; ignoring clean accuracy; not considering adaptive attacks.

    Origin & history

    Documented papers are Szegedy et al. (2013/2014), Goodfellow et al. (2014/2015) and Madry et al. (Preprint 2017, ICLR 2018).

    Use in mid-sized companies

    • Potential applications: Red teaming and robustness tests for classification, document or image recognition systems before productive use.

    Related terms

    Matching Flow-alerta service

    AI opportunity analysis & roadmapAI consulting: AI potential analysis & roadmap on tasks, data and prerequisites of a possible first AI use case.

    Matching tool / use case

    View use casesUse Case: Robustness test of an ML system; Tools: Adversarial Robustness Toolbox, Foolbox, own test suites

    Sources

    • arxiv.orgGoodfellow, Shlens & Szegedy (2014), „Explaining and Harnessing Adversarial Examples“
    • arxiv.orgSzegedy et al. (2013), „Intriguing properties of neural networks“
    • arxiv.orgMadry et al. (2017 / ICLR 2018), Towards Deep Learning Models Resistant to Adversarial Attacks

    Quellenprüfung: 09.10.2026. Anwendungsbeispiele sind mögliche Szenarien, keine behaupteten Kundenfälle.

    Editorially reviewed on:

    Next step

    Put knowledge into practice

    Discuss your projectView use cases