Published: 22 July 2025
Facing Up to AI Security
š§ Securing the Future: A CISOās Guide to the UK AI Cyber Security Code of Practice
As AI becomes embedded in everything from finance to national infrastructure, the risks it introducesāboth to and through cyber systemsāare evolving at pace. To help manage these risks, the UK government recently published the AI Cyber Security Code of Practice, a voluntary guide aimed at improving how AI systems are developed, deployed, and defended.
But what does this mean for CISOs and cybersecurity teams? How useful is the Code in real-world applications, and where might it fall short?
š What Is the AI Cyber Security Code of Practice?
The UKās AI Cyber Security Code of Practice is a voluntary framework designed to help developers and operators secure AI systems throughout their lifecycle. It aligns with existing cybersecurity standards (like NIST and ISO 27001) while addressing emerging risks unique to AI technologies.
The Code is built around five core principles:
- Secure AI Development
- Secure AI Deployment
- Secure Supply Chain
- Technical and Organisational Governance
- Continuous Risk Management
ā Benefits for CISOs and Security Leaders
1. Structured Security Framework
The Code provides a clear, repeatable structure for managing AI-specific cyber risks and integrating AI security into broader enterprise strategies.
2. Supply Chain Risk Awareness
It emphasizes the importance of vetting and securing third-party AI components, APIs, and training datasets.
3. Governance and Risk Ownership
Encourages defined roles and responsibilities for AI security, improving cross-team governance.
4. Continuous Threat Monitoring
Promotes real-time model monitoring to detect drift, bias, or adversarial threats early.
ā ļø Key Challenges in Implementation
1. Voluntary, Not Mandatory
Without enforcement, industry adoption may be inconsistentācreating gaps in security coverage.
2. Fast-Changing Threat Landscape
New AI-specific threats like prompt injection and model theft evolve faster than guidance can be updated.
3. Complex AI Supply Chains
Deeply nested and often opaque dependencies present difficulties in auditing and securing end-to-end systems.
4. Skills and Resourcing Gaps
Many organisations lack access to skilled personnel for AI threat modeling or red teaming.
š§Ø Where the Code May Fall Short
1. Limited Focus on Malicious AI Use
The Code focuses on protecting AI, but not enough on preventing AI misuse (e.g., deepfakes, autonomous phishing).
2. Neglect of Extended Attack Surfaces
Edge AI (in IoT, vehicles, etc.) expands risks the Code doesnāt fully address.
3. Gaps in Data Provenance and Model Drift
Data lineage and ongoing validation remain weak points despite the emphasis on secure training inputs.
4. Consequences of Getting It Wrong
Failures can lead to major breaches, regulatory penalties, safety risks, and loss of public trust in AI systems.
š Final Thoughts: Use It, But Donāt Stop There
The UK Code is a strong starting point, not a complete solution. CISOs should treat it as a baseline while supplementing it with:
- AI-specific red teaming and penetration testing
- Third-party audits of models and training pipelines
- Governance integration with GRC and regulatory compliance
- Lifecycle risk assessments and supply chain vetting
