Strategy
AI security budget 2027: how much to spend on securing AI
AI was just 3% of the average security budget in 2026, and 69% of CISOs expect to spend more. Here's what the money buys, what it costs, and how to size the line.
For most organizations, AI security is still a small line. AI made up 3% of the average security budget in 2026, the first year IANS tracked it as its own category. It’s the line most CISOs plan to grow, though: 69% expect AI-related security spending to rise in the next budget cycle, up from 46% whose AI security spending rose in the last one. Overall, 45% of CISOs expect an increase of 1–10% and 24% expect more than 10% (IANS, September 2026).
The market numbers show the same trend. Gartner forecasts the market for securing AI will grow from $2.8 billion in 2026 to $4.8 billion in 2027, up 68.7% (Gartner, August 2026).
Two different AI budgets
“AI security” covers two kinds of spending with different owners, vendors and risks. Budget them separately:
| AI for security | Security for AI | |
|---|---|---|
| What it is | AI inside your security tools: SOC copilots, AI triage, automated investigation | Controls that protect your own AI use: employee gen-AI apps, copilots, agents and models you build |
| Typical buyer | SecOps | Security architecture, data protection, AppSec |
| How it’s priced | Often by usage (compute units, tokens) on top of a platform license | Per user or per app, plus gateway traffic |
| 2026 signal | 69% of CISOs named it as a destination for net-new money, more than any other area (IANS) | Half of security leaders say attacks on AI systems are among the threats they’re least prepared for (PwC) |
What the securing-AI market is buying
Gartner splits the market into four named segments plus “other”. Here is its forecast in millions of US dollars:
| Segment | 2026 | 2027 | 2027 growth |
|---|---|---|---|
| AI application security | $508M | $851M | 67.5% |
| AI usage control | $433M | $749M | 73.0% |
| AI governance platforms | $275M | $462M | 68.0% |
| AI gateways | $251M | $429M | 70.9% |
| Other securing AI | $1,368M | $2,292M | 67.5% |
| Total | $2,835M | $4,783M | 68.7% |
Source: Gartner, August 26, 2026. Gartner’s release doesn’t say what “other securing AI” includes, although it’s nearly half the total. Gartner expects the whole market to reach almost $7.7 billion by 2028.
In budget terms:
- AI usage control governs which gen-AI apps employees use and what data goes into them. It’s the fastest-growing segment, and usually the first purchase for organizations that don’t build their own AI.
- AI application security protects AI apps and agents you build or deploy, for example against prompt injection, data leakage and unsafe tool use. You need it if you build AI apps or agents.
- AI gateways sit between your users or apps and the models, enforcing policy and logging usage.
- AI governance platforms inventory models and use cases and track them against policy and regulation.
Gartner predicts that by 2029, over half of successful attacks on AI agents will exploit access control weaknesses and prompt injection. So part of the “security for AI” budget is really identity work: giving agents their own identities, scoped permissions and logging.
What AI security actually costs
Most AI security vendors are quote-only. The clearest public price list is Microsoft’s, for AI in the SOC:
- Microsoft Security Copilot (standalone): $4 per provisioned security compute unit (SCU) per hour. You must provision at least one SCU, billed monthly. Usage beyond what you provision costs $6 per overage SCU, up to a cap you set. One SCU provisioned around the clock is about $2,920 a month, or roughly $35,000 a year.
- Included with Microsoft 365 E5 and E7: 400 SCUs a month for every 1,000 paid user licenses, up to 10,000 SCUs a month, at no extra cost. A 400-license tenant gets 160 SCUs a month. For scale, one SCU provisioned around the clock covers about 730 SCUs of usage a month, so 400 included SCUs is a little over half of one always-on unit (our arithmetic from Microsoft’s billing rules). Microsoft says it will eventually throttle usage above the allocation and let customers buy more at $6 per SCU, with 30 days’ notice.
That’s the cost pattern Forrester calls token economics, which its 2027 planning guides name as a new security budget line item (SecurityBrief). The license is only part of the cost. Agents and copilots also consume metered compute, so set a usage line, a cap and an owner.
Who pays for it
IANS found that AI spending is tracked and funded three ways:
- 24% of organizations fund it as a separate line item or subcategory.
- 38% fund it from within the broader security budget.
- 38% fund it from outside the security budget, typically from IT, data or innovation budgets.
That last group matters for benchmarking. The 3% average counts only AI spending inside the security budget, so total spending on AI security is likely higher than the share suggests.
When AI security is funded from someone else’s budget, the CISO owns the risk without the money. Before the budget locks, agree with the teams running AI projects which security controls come out of their project budgets and which come out of yours.
How to size your AI security line
- Inventory AI use first. List sanctioned gen-AI apps, copilots, agents and any models you build or fine-tune. The size of the line depends mostly on whether you build AI or only consume it.
- Extend the controls you already pay for. DLP, CASB or SSE, and identity governance can often cover AI apps with policy changes and no new license. DLP typically runs about $3–$15 per user per month.
- Add usage control or a gateway when gen-AI use is broad and sensitive data is involved.
- Add AI application security only if you ship AI apps or agents to employees or customers.
- Pilot AI for security on metered capacity. Start with one provisioned unit or your included allocation, measure analyst time saved, and set an overage cap before scaling.
Split the line by owner as you go: AI-for-security tools usually sit with SecOps, and controls for employee and product AI with data protection or AppSec. IANS doesn’t report how the 3% divides between the two.
As a reference point, take the $2.3 million budget from our 2027 planning example. The 2026 average AI share of 3% is $69,000, a little less than two Security Copilot SCUs provisioned around the clock. If you have no AI line today, starting at the 2026 average and adding to it once you can show measured results is a defensible ask. That’s our planning rule of thumb, not a benchmark. Large increases tend to come from organizations already spending: IANS found that 75% of those that raised AI security spending by more than 10% last cycle expect to do so again.
Where AI security fits in the 2027 budget
IANS also found that organizations planning big AI security increases were more likely to have budget growth overall: 45% of those planning increases above 10% saw their total security budget grow by more than 5% this cycle, compared with 12% of those planning no AI-specific security spending. That’s a correlation, not proof that an AI line wins budget. But an AI risk tied to a named control does make a concrete case.
For the rest of the 2027 plan, including staff, renewals and a worked example, see our 2027 cybersecurity budget benchmarks. To estimate a full budget for your size and industry, run the free cybersecurity budget calculator.
Prices and figures are as published by each source when we checked them in November 2026. Microsoft notes that its prices are estimates and vary by agreement and currency. Survey percentages describe respondents, not every organization.
References
- IANS / Artico 2025 Security Budget Benchmark (opens in a new tab) · verified Aug 2026
- DLP pricing — Microsoft Purview, Forcepoint, Zscaler, Symantec (opens in a new tab) · verified Jun 2026
Put this into a number
Build a sourced, defensible budget from your own risk profile — free and ungated.