How to Integrate AI Risk into Your Existing Cybersecurity Programme
Artificial intelligence is no longer a technology of the future—it is here, and it is already changing how your business operates. From AI assistants helping your teams to machine learning models optimising supply chains, AI is quickly becoming a core part of the modern enterprise. For many Chief Information Security Officers (CISOs), the first instinct is to create a separate security strategy just for AI.
But what if this approach, while seeming logical, is a mistake? What if building a dedicated "AI Security" silo creates more risk than it solves?
The most effective and resilient way to manage AI risk is not to build new, separate defences. It is to expand the boundaries of your current cybersecurity programme to include AI. It’s time to normalize AI security, treating it not as a special new problem, but as another powerful technology that must be governed by your existing, mature security frameworks.
The Temptation of the Silo—And Why It Fails
The idea of creating a specific "AI Security Programme" is easy to understand. AI feels new, the risks seem different, and unique regulations are on the horizon. A 2025 survey found that 66% of organisations believe that managing GenAI risks requires significant changes to their cybersecurity risk management approach1. This often leads to creating AI-specific policies, workflows, and teams.
While this approach gives a clear project scope for the short term, it is a long-term strategic trap. AI is becoming deeply integrated into almost every software and service your organisation uses. Trying to manage AI security in isolation is like trying to manage "internet security" as a separate function from your main security operations—it is impossible and inefficient. A stand-alone AI security programme will quickly become a bottleneck, creating unmanageable complexity and cost as the technology continues to evolve at high speed.
The Litmus Test: Is This Truly a New Cybersecurity Risk?
Before you draft a new "AI Acceptable Use Policy," you must differentiate between new AI risks and existing business risks that now involve AI. A CISO's responsibilities have clear boundaries focused on cybersecurity.
Use this simple litmus test to decide if a risk belongs to your team: Is there already a rule for this type of risk in your existing cybersecurity policies?
Ethical Use and Content Validity: Is your security team responsible for the ethical implications of how an employee uses a spreadsheet? No. That is a broader business or HR issue. Similarly, the ethical use of AI and the validity of its output should be managed by a cross-functional governance committee, not just the CISO.
Data Exposure and Leakage: Is your team responsible for stopping employees from uploading confidential files to their personal cloud accounts? Yes. Therefore, stopping employees from pasting sensitive corporate data into a public AI tool is a task for your existing
Data Loss Prevention (DLP) and SaaS security policies. The challenge of "Shadow AI" is simply the next evolution of "Shadow IT."
Vulnerable Software: Is your team responsible for scanning vulnerabilities in the software your developers build? Yes. Then the AI models, libraries, and components they use must be included in your existing Secure Software Development Lifecycle (SDLC) practices.
The key realisation is this: the majority of AI-related cyber risks are not new. They are new versions of existing threats that your current security programme is already designed to address.
Your 3-Step Plan for Normalizing AI Risk
Instead of building a new programme from zero, CISOs should focus on a strategic mission to absorb AI into their current security framework. This approach is more efficient, scalable, and ultimately, more secure.
Step 1: Deconstruct AI and Map Your Existing Controls
For a moment, ignore the "AI" label. Break down any AI system into its basic technical parts:
data, code, and infrastructure.
Data: AI systems use data for training, processing, and generating outputs. Your current data governance policies—covering data classification, handling, and access management—are your first and most important line of defence. Your task is to expand the scope of these policies to officially include AI applications.
Code: AI models and applications are software. Your existing SDLC and DevSecOps standards for code scanning, quality testing, and vulnerability management apply directly to AI development.
Infrastructure: AI runs on infrastructure, whether in your data centre or in the cloud. Your existing controls for platform, network, and vulnerability management are all relevant.
When you map your current controls to these components, you will likely find that 80% to 90% of the cybersecurity risks are, in principle, already covered. The result of this exercise is a gap analysis, showing you the few truly new risks that require your special attention.
Step 2: Expand Your Governance to Fill the Gaps
Your gap analysis will highlight the few genuinely new risks that your existing policies might not cover. These often involve new attack surfaces like:
Prompt Injection: Crafting malicious inputs to trick an AI model into performing unintended actions.
Model Poisoning: Intentionally corrupting an AI model's training data to compromise its integrity and security.
For these unique risks, you should expand your governance. However, this does not mean you need a completely new policy document. Instead, amend your existing policies. Add a section on prompt injection to your secure coding standards. Update your third-party risk assessment to check for model poisoning. The goal is integration, not isolation.
Step 3: Set a Deadline for "AI Security" to Disappear
The final objective is to make the term "AI Security" unnecessary. In a future where AI is everywhere, all cybersecurity must be AI-aware. While a temporary, focused effort on AI risks is needed now to build skills and close immediate gaps, this should not be a permanent strategy.
Create a strategic 18-month plan to fully absorb all AI-specific security governance into your standard, day-to-day cybersecurity operations9. By the end of this period, your data security policy should naturally cover data in AI systems, and your incident response plan should be ready for an AI-related breach as a standard scenario.
The Key Insight: You Are More Prepared Than You Think
The constant hype and news about AI can make even experienced security leaders feel they are starting from zero, facing a threat so new that their existing programmes are no longer valid.
This is simply not true.
Your organisation's cybersecurity framework, built on years of experience in data protection, threat management, and secure architecture, is the strong foundation you need. By normalizing AI security, you build upon this foundation, avoid creating inefficient silos, and develop a security programme that is truly ready for the future. The mission is not to build a new security programme for AI; it is to make your current, proven programme AI-aware. That is a much smarter, and far more effective, goal.
