ISO/IEC 27090: A New Standard Addresses Cybersecurity Challenges in AI
Nov 05, 2026Audit / Digital TrustArtificial intelligence creates new opportunities for organizations, but it also introduces cybersecurity risks that they need to understand. A new international standard, ISO/IEC 27090, provides guidance on identifying and addressing security threats specific to AI systems, complementing existing information security practices.
As organizations increasingly integrate AI into their operations, protecting this technology involves more than securing the infrastructure on which it runs. The data used to train AI models, the models themselves and the way systems respond to inputs can all present potential security challenges.
Published in 2026, the standard ISO/IEC 27090, Cybersecurity – Artificial intelligence (AI) – Addressing security threats and compromises to AI systems, helps organizations understand these challenges and consider appropriate measures throughout the AI lifecycle.
Understanding the Security Risks of AI
Traditional cybersecurity measures remain essential, but AI systems can be exposed to threats that require additional consideration.
For example, manipulated training data can influence how a model learns and behaves, an attack known as data poisoning. Repeated queries may be used to replicate a model, known as model extraction, while other attacks exploit vulnerabilities in the way certain inputs are processed. Depending on the system, attackers may also attempt to infer sensitive information from model outputs.
These risks can affect organizations in different ways, with implications for the protection of intellectual property and confidential information and the reliability of AI-supported business processes. Their significance depends on the technology being used, its application and the potential consequences of a compromised system.
The scope is not limited to generative AI or chatbots: ISO/IEC 27090 addresses security threats to AI systems in general. Key aspects of the standard include understanding these threats and identifying technical and organizational measures that can help prevent, detect and address them. Relevant activities include threat modelling, which identifies potential attack scenarios, and red teaming, which tests defenses through simulated attacks.
Complementing Established Information Security Practices
The new standard builds on existing cybersecurity principles rather than replacing them. Organizations can use it to assess how AI-specific risks fit into their current security arrangements, including risk assessments, access controls, testing and monitoring.
ISO/IEC 27090 also complements two established management system standards: ISO/IEC 27001, which specifies requirements for an Information Security Management System (ISMS), and ISO/IEC 42001, which specifies requirements for an Artificial Intelligence Management System (AIMS).
While ISO/IEC 27001 provides a framework for managing information security risks and ISO/IEC 42001 addresses the overall management of AI, ISO/IEC 27090 focuses specifically on cybersecurity threats and compromises affecting AI systems. It also draws on established information security practices, including guidance such as ISO/IEC 27002.
Organizations can use ISO/IEC 27090 alongside their existing management systems to address AI-specific security risks. Because it provides guidance rather than certification requirements, it supports those systems without replacing their requirements.
What Does This Mean for Businesses?
For organizations developing, deploying or using AI, the standard provides an opportunity to assess how security can be addressed throughout the entire AI lifecycle.
A practical starting point is to identify which AI systems are in use, assess the data and models involved, and review the security of third-party solutions. Organizations can then determine how potential vulnerabilities will be identified and managed. Information security, AI development, risk management and business teams can work together to integrate these activities into broader organizational processes.
The relevance of these measures extends across all industries, particularly where AI supports important business decisions, processes sensitive information or forms part of critical operational activities.
As AI adoption continues to grow, organizations need to keep their security practices aligned with the technologies they use. ISO/IEC 27090 provides additional guidance to support these efforts, helping businesses consider AI-specific threats alongside their established cybersecurity measures.
For organizations already working with information security and AI management systems, the standard offers a useful reference for developing a more comprehensive approach to AI security.
FAQ
ISO/IEC 27090:2026 is a guidance standard that addresses security threats and compromises specific to AI systems throughout their lifecycle.
No. It provides guidance rather than requirements for certification or an AI management system.
It is useful for cybersecurity teams, AI developers, data managers and those responsible for procuring or overseeing AI systems. It covers AI systems broadly, not just generative AI or chatbots.
- ISO/IEC 27001 specifies requirements for an Information Security Management System.
- ISO/IEC 42001 specifies requirements for an Artificial Intelligence Management System.
- ISO/IEC 27090 focuses specifically on cybersecurity threats and compromises affecting AI systems.
No. It complements established cybersecurity practices by providing additional security guidance specific to AI.
ISO/IEC 27090 addresses a range of security threats affecting AI systems, including data poisoning, model theft, adversarial attacks and vulnerabilities in the way AI systems process inputs and generate outputs. It provides guidance to help organizations understand these threats and consider appropriate technical and organizational measures to prevent, detect and address potential security compromises.
ISO/IEC 27090 is not limited to generative AI, large language models or chatbots: it addresses cybersecurity threats affecting AI systems broadly. Its guidance is relevant to organizations developing, deploying, integrating or using AI technologies across different industries and applications. The relevance of individual threats and measures depends on the system and its use.
Organizations can use ISO/IEC 27090 to assess AI-specific cybersecurity risks throughout the AI lifecycle. This includes identifying potential threats, evaluating the security of AI models and training data, reviewing third-party AI solutions and considering appropriate controls. The guidance can also complement existing information security and AI management practices under ISO/IEC 27001 and ISO/IEC 42001, helping organizations develop a more comprehensive approach to AI cybersecurity.
