Why Security Fundamentals Still Matter for AI
Every few years, a technology arrives that tempts the security community to throw out the playbook ...
Organisations and governments are racing to adopt AI faster than they can secure and govern it. The evidence for that gap, and what it takes to close it.
Organisations and governments want AI everywhere, and fast. Their ability to secure and govern it is not keeping up. That gap between ambition and ability to govern is where the real risk now accumulates; this is the reason every organisation adopting AI needs a deliberate AI Safety and Security function.
The World Economic Forum puts it plainly: organisations are embracing AI at scale “even as governance frameworks and human expertise struggle to keep pace” (WEF Global Cybersecurity Outlook 2026). The financial sector, usually among the best defended, says the same. FS-ISAC, the banks’ own threat-sharing body, warns that frontier AI capabilities “have progressed faster than many organizations have been able to adapt” (FS-ISAC).
Worldwide AI spending is forecast to reach about $2.52 trillion in 2026 (Gartner). Spending on securing AI is forecast at about $2.8 billion in the same year (Gartner via IT Brief). That is roughly one dollar spent on securing AI for every thousand dollars spent on adopting it.
Security spend is growing quickly, but from a very low base, and it is not where the gap is widest. The deeper problem is that the organisations deploying AI have not built the ownership, processes and skills to run it safely.
The evidence across governance, identity, visibility and skills tells a consistent story.
| Where the gap shows | What the evidence says | Source |
|---|---|---|
| Governance | 68% of breached organisations had no AI governance, up from 63% a year earlier. Five of six governance controls lost adoption; approval processes for AI deployments fell from 45% to 38% | IBM Cost of a Data Breach 2026 |
| Basic access control | 92% of organisations with an AI-related breach lacked proper AI access controls; only 40% apply access controls to AI models and data | IBM X-Force |
| Shadow AI | Unsanctioned AI was involved in 43% of incidents, up from 20%; those breaches averaged $5.39 million | IBM 2026 |
| Ownership | Only 19% coordinate AI governance and security teams; 20% leave agent management to whichever team deployed them | IBM 2026, Okta CISO Insights |
| Board alignment | Only 31% of security leaders are fully aligned with their board on acceptable AI risk | Okta CISO Insights |
| Agent visibility | 47% can identify all their AI agents, 46% can control what they access, 45% can authorise what they do | Okta CISO Insights |
| Perception versus reality | 68% claimed strong visibility of their agents, yet 82% found previously unknown agents in the past year | CSA and Token Security |
| Detection and containment | 95% doubt they could detect malicious activity by an AI agent; 5% feel confident they could contain a compromised AI system | Cybersecurity Insiders 2026 |
| Skills | 95% of UK security teams report at least one skills gap, 58% critical or significant; AI is the most urgent gap, named by 42% | ISC2 UK findings |
| Threat tempo | AI-driven attacks rose 56% year on year, adding about $1 million to the cost of a breach; the average breach now costs $4.99 million | IBM X-Force |
Some of these surveys were run or sponsored by companies that sell security products, so read individual figures as indicators not necessarily the full facts. The pattern matters more than any single number, and the pattern is the same across every source: adoption and attacker tempo are compounding faster than governance, identity and skills.
It would be easy to read this as a failure of security technology. It’s not as simple as that. Most of the controls on the market broadly do what they claim. However, the key real-world constraint is the operating model around them: who owns AI risk, how fast assurance can move, and whether there are people able to run the controls to address the risks.
The heart of it is a clock-speed mismatch:
The organisation that has to have the ability to absorb the new technologies. When the organisation cannot, the tool becomes another console nobody watches, and the risk is not addressed. The gap is therefore less about buying capability and more about building the capacity to use it.
The gap is an organisational problem, so the answer is an organisational one. It is not a single product or a policy document; it is a function with the mandate, the people and the operating rhythm to keep pace with AI adoption. That is what the rest of this guide sets out to help you build.
Security fundamentals endure. What has changed is the pace at which they must now be applied. The organisations that thrive with AI will be the ones that build the capacity to govern it as deliberately as they build the capacity to adopt it.
Continue building your AI Safety and Security function.
Start with the FoundationsEvery few years, a technology arrives that tempts the security community to throw out the playbook ...
Prompt injection has rapidly become the most discussed vulnerability class in AI security, and for ...
The EU AI Act represents the most comprehensive attempt by any jurisdiction to regulate artificial ...