There is a conversation that needs to happen in every enterprise running an AI programme, and in most organisations it isn't happening well. It's the conversation between the CIO and the CISO.
These two roles have always had some creative tension: the CIO driving adoption and velocity, the CISO managing risk and protecting the organisation. That tension is productive when both parties understand each other's constraints and share a common framework for making decisions. AI has added significant new complexity to that relationship, new risks, new attack surfaces, new governance requirements, and in many organisations the existing ways of working between these two functions aren't keeping pace.
Why AI specifically strains this relationship
AI systems have opaque decision logic
Traditional software does what it is programmed to do. AI systems learn from data and produce outputs that are often difficult to fully explain or predict. This creates explainability and auditability requirements that most security frameworks weren't designed to address.
AI expands the attack surface in novel ways
Prompt injection, model poisoning, adversarial inputs, training data leakage, these are attack vectors that most security programmes are still developing competency in. The CISO who last reviewed their threat model before the current generation of generative AI has a gap to close.
AI creates new categories of data risk
AI systems often require access to large volumes of sensitive data for training, fine-tuning, or inference. The data flows involved, ingestion, labelling, training, deployment, create multiple new exposure points that need to be secured and governed.
The pace of AI deployment outstrips security review capacity
Business units and technology teams are under pressure to ship AI quickly. Security review processes designed for traditional software releases struggle to scale to the velocity of AI deployment. The result: security reviews either become bottlenecks that slow deployment, or teams start working around them.
The CIO wants to move. The CISO wants to protect. In the best organisations, these are the same goal expressed differently. In struggling organisations, they become opposing forces that stall AI programmes or create unacceptable risk.
The four fault lines
In my experience advising enterprise leaders across the region, four specific fault lines account for the majority of CIO-CISO friction in AI programmes:
- Risk appetite misalignment: The CIO and CISO have different implicit thresholds for acceptable AI risk, and they've never explicitly agreed on what the organisation's actual risk appetite should be.
- Governance process friction: Security review processes aren't designed for AI's development cadence, creating structural delay that breeds pressure to bypass.
- Data access disputes: AI programmes need data that security policy restricts. The resulting disputes about data access slow programmes and create adversarial dynamics.
- Shadow AI: Employees are using AI tools, many of them cloud-based, consumer-grade, without IT or security awareness. The CISO discovers this during an incident rather than proactively.
What good CIO-CISO collaboration on AI looks like
A shared AI risk appetite, agreed at board level
The most effective starting point is a joint conversation, ideally facilitated by the board or CEO, to agree on the organisation's AI risk appetite. Not in the abstract, but in specific terms: what categories of AI use case are we willing to deploy with what level of oversight? What are our non-negotiable security requirements, and what are the areas where we accept more risk in exchange for more speed?
Security as a design requirement, not a gate
The security function needs to move from reviewing AI systems at deployment to helping design them from the start. This requires embedding security thinking into the AI development lifecycle: threat modelling for AI systems, security requirements defined at the use case stage, and regular security reviews throughout development rather than one gate at the end.
A joint AI security and governance framework
Rather than the CIO and CISO operating under separate frameworks that occasionally conflict, the most effective organisations build a single AI security and governance framework that addresses both the deployment velocity the CIO needs and the risk management the CISO requires.
Regular joint reviews of the AI risk landscape
The AI security landscape changes rapidly. New attack vectors emerge. Regulations evolve. The threat model for a model deployed 18 months ago may look quite different today. CIO-CISO joint reviews of the AI risk landscape, at least quarterly, help both functions stay aligned and identify emerging issues before they become incidents.
A shared position on shadow AI
Rather than banning shadow AI use, which rarely works, the most effective approach is to jointly develop a clear policy on personal AI tool use, train staff on what is and isn't acceptable, and create an approved channel for teams that want to use AI tools for legitimate purposes.
The conversation to have this quarter
If you are a CIO or CISO reading this and you don't have a regular joint conversation with your counterpart about AI, that is the most important thing you can fix in the next 90 days. You don't need a new framework to start. You need 90 minutes in a room, ideally with the CEO present, to align on risk appetite and agree on how you're going to work together on AI.
AI programmes don't stall because of bad technology. They stall because of bad alignment. The CIO-CISO relationship is one of the most important alignment problems in enterprise AI, and one of the most fixable.
Vijay Jaswal is Founder and CEO of Kudo Advisory. Reach him at info@kudoadvisory.com or on LinkedIn.