What is BS ISO/IEC 27090:2026 – Addressing security threats and compromises to AI systems about?
Understand how AI systems get attacked, and what to do about it.
BS ISO/IEC 27090:2026 gives you guidance on the security threats and compromises specific to AI systems. It shows how attacks and vulnerabilities can affect AI data, models, prompts, outputs and services across the lifecycle, the consequences that follow, and approaches to detection and mitigation.
It complements your existing practice. Use it to identify AI-specific risks and safeguards that support your ISO/IEC 27001 information security and ISO/IEC 42001 AI management frameworks.
Who is BS ISO/IEC 27090:2026 – Addressing security threats and compromises to AI systems for?
Organizations of all types and sizes that develop, deploy or use AI, including:
- CISOs, security leaders and information-security managers;
- AI governance, risk, compliance and audit teams;
- AI/ML developers, data scientists and software engineers;
- security architects, threat modellers, red teams and incident responders; and
- procurement and supplier-assurance teams.
What does BS ISO/IEC 27090:2026 – Addressing security threats and compromises to AI systems cover?
-
Assets: AI data, models, prompts, outputs and services.
-
Attacks on training and models: data poisoning, evasion and direct model poisoning.
-
Attacks on privacy and IP: membership inference, model inversion, model exfiltration and direct model theft.
-
Attacks on prompts and outputs: prompt injection and output-based injection.
-
Leakage: training-data and model-input leakage, and sensitive model outputs.
- Response: detection and mitigation considerations across design, development, deployment and operation.
How can BS ISO/IEC 27090:2026 – Addressing security threats and compromises to AI systems help you?
- Identify AI-specific security threats and the assets they affect.
- Understand the consequences for confidentiality, integrity and availability.
- Choose proportionate approaches to detection and mitigation.
- Give security, AI, risk, governance and procurement teams a common language.
- Apply a whole-life-cycle approach to AI security.
- Connect AI safeguards to your information-security and AI-management approach.