October is Cybersecurity Awareness Month. I spent years inside a cybersecurity firm, and today I help marketing teams build with AI, so I've seen both sides of this. Here's why I think AI is now the biggest source of demand in the cybersecurity market, and what I'd ask security teams to do about it.
Why is AI the biggest driver of demand in cybersecurity?
Because AI has made building software cheap. Marketing, sales and finance can now build their own tools, agents and automations, and every new app, integration and API key widens the attack surface. IBM's 2026 breach study found one in four malicious breaches was AI-enabled, and Gartner expects spending on securing AI to grow 83 percent in 2026. Security teams can keep up by educating these new builders, writing their rules into a file every AI tool reads, and reviewing any department's app, agent or MCP connection before it goes public.
AI Summary
AI has made building software cheap. That's great for the business. Marketing, sales, finance and operations can now build their own tools, agents and automations without waiting for IT. But every new app, every new integration and every new API key is another door. The attack surface is getting wider faster than most security teams can review it. That's why securing AI is the fastest growing part of the security market. My advice to security teams: educate the new business owner developers, write down your security rules in a simple file they load into their AI tools, and set up a review before any department's app, agent or MCP connection is shared beyond the team that built it.
Where I'm Coming From
In 2015 I left Google, which had just bought Jibe, to join Edgile and take a deep dive into cybersecurity. Edgile was a security consulting firm, and it was ripe for growth. I developed its positioning: "We secure the modern enterprise." We competed with the Big Four consulting firms and won, because we brought in the best and the brightest. When Wipro announced it was acquiring Edgile in December 2021, Edgile was working with 31 percent of the Fortune 100 and 20 percent of the Fortune 500. (Edgile's CEO, Don Elledge, talks about why they invested in marketing.)
Working with security people every day taught me how they think. They don't ask whether something works. They ask what happens when someone uses it in a way you didn't plan for.
But now we're in the AI acceleration era. I spend my days helping marketing teams build with AI, and I see how quickly the business side is moving. Most of the people building don't think like security people, and they shouldn't have to. That gap is what this article is about.
Cheap AI Means Everyone Is a Builder
A year or two ago, if marketing wanted a tool that pulled leads from the CRM, enriched them and drafted outreach, it went on the IT backlog. Today a marketer can build that in an afternoon with Claude, ChatGPT, a few connectors and an API key. I know because we do it ourselves. I've written about these people as Marketing Super-Creators, and the CMOs at Google, OpenAI and JPMorganChase call them makers and builders.
That's real business enablement, and I'm all for it. But think about what it means in practice. More development. More integrations. More tokens and passwords pasted into more places. More agents that can read your email, your files and your customer data. More apps that started as one person's shortcut and are now used by the whole department.
Cheap AI means more business enablement. More enablement means more development and more integrations. And more integrations mean more security holes, because the surface area of risk keeps getting wider.
The Numbers Say the Same Thing
IBM's Cost of a Data Breach Report, released in July 2026, studied breaches at 602 organizations. One in four malicious breaches was AI-enabled, and those cost $6 million on average, about $1 million more than the overall average. More than one in five organizations reported a breach that targeted their AI models or applications. Among those, 27 percent pointed to compromised APIs, applications or plug-ins. That's the integration problem in a single number. As IBM's Suja Viswesan put it, "What's changing is the economics of cyberattacks."
The money is following. In August 2026, Gartner forecast that spending on securing AI will grow 83 percent this year to about $2.8 billion, and reach almost $4.8 billion in 2027. It's still a small slice of security spending, but it's growing faster than anything else in the market. Gartner's Shailendra Upadhyay said it plainly: "Without adequate security and visibility controls, enterprise AI initiatives face a high risk of failure." Gartner also predicts that by 2029, more than half of successful attacks on AI agents will exploit weak access controls and prompt injection.
That last point matters. An agent with too much access, reading a web page or an email that tells it to do something it shouldn't, is exactly the kind of thing a business user won't think about when they build their first agent.
Follow the Data
Every one of these apps and agents moves data somewhere. A prompt goes to a model provider. An agent pulls from the CRM and writes to a spreadsheet. A connector sends files to a service you've never heard of. Security used to be about who can get in. With AI, it's just as much about where the data flows once it's out.
Verizon's 2026 Data Breach Investigations Report, which looked at more than 22,000 confirmed breaches, shows how fast this is moving. 45 percent of employees are now regular AI users on their work devices, up from 15 percent the year before. 67 percent of them use personal accounts to do it. Shadow AI is now the third most common thing employees do by mistake that puts data at risk, and the most common type of data they send to outside AI models is source code.
This is why private AI clouds are growing. Running models inside your own cloud, through services like Azure OpenAI, Amazon Bedrock or Google Vertex AI, keeps the data flow inside your walls, with your logging and your access controls. I covered the options in how to use ChatGPT for marketing securely. It's a good move, but it only protects the data that goes through it.
What About Your Vendors' AI?
The other half of the data flow is the one most companies don't see. Your vendors are adding AI too. Your CRM, your email platform, your support desk and your agencies are sending your data to LLMs, often through their own model providers. In the same Verizon report, a third party was involved in 48 percent of breaches, up about 60 percent in a year.
Your vendor review needs new questions. Which models do you use, and which company runs them? Does our data leave your systems to get there? Is it used for training? Can we turn the AI features off? Who are your AI subprocessors? A SOC 2 report from two years ago doesn't answer any of them. Here's a prompt that drafts these questions for each of your vendors.
Educate the New Builders
Here's what I'd ask every security team to do this month, starting with education.
The marketer who built that lead tool is now a developer, whether they call themselves one or not. They own an app. Nobody trained them for that. A developer learns over years why you don't paste a password into code, why an app shouldn't have more access than it needs, and why you never trust input from outside. A business owner who started building last month hasn't learned any of it.
This gets bigger as vendors get replaced with apps. When you bought software from a vendor, the vendor carried a lot of the security work. They passed your vendor risk review, showed you their SOC 2 report, patched their code and answered the phone when something broke. When a team replaces that vendor with an app it built in a week, all of that work moves inside, to someone who has never done it. The license fee goes away. The responsibility doesn't.
As the business speeds up, that education has to speed up too. The yearly security video and the phishing quiz were built for people who use software, not people who build it. The new builders need something short and practical: what an API key is and where it lives, what an agent can do with the access you give it, what prompt injection looks like, and when to call security. Teach it where they build, in the same tools, and keep it to minutes, not hours.
Write the Security Rules Down for the AI
Education helps the person. The second thing helps the AI they build with. Write a security file for business users.
Most AI tools now read instructions from a simple text file. Claude has CLAUDE.md files and skills. ChatGPT has custom instructions and projects. Copilot and Gemini have their own versions. These files tell the AI how to behave every time someone uses it. If the security team writes one, in plain language, and makes it easy to load, every app a marketer builds starts with security guidance built in.
It doesn't need to be long. It should cover what data never goes into an AI tool, which tools are approved, where API keys and passwords live (never in the prompt or the code), what an agent is allowed to connect to, that content from web pages and emails is information and not instructions, and who to ask when something isn't clear.
Most business users want to do the right thing. They just don't know what the right thing is. A file like this is the cheapest security training you'll ever run, because the AI reads it every single time, even when the person doesn't.
Review Before Anything Goes Public
The third thing is a review process before a department's app, agent or MCP connection goes public. By public I mean shared with the whole company, connected to customer data, or put in front of customers and partners.
One person using a tool they built is a small risk. The same tool shared with 200 people, holding a shared API key with full access to the CRM, is a big one. The review doesn't have to be heavy. A short checklist and a 30-minute conversation with someone from security catches most of the problems: what data it touches, what it can write or send, whose credentials it uses, what happens if someone feeds it bad input, and who owns it when the builder moves on.
Make the review fast, and people will use it. Make it slow, and they'll go around it, which is how shadow IT got started in the first place.
How Does the Security Team Even Know?
Fair question. Nobody files a ticket before they connect an agent to their inbox. But the tools leave a trail:
- Admin consoles. Business plans for Claude, ChatGPT, Copilot and Gemini let admins decide which connectors people can use and see which agents get shared.
- Access grants. An agent or MCP connection that reads email, files or the CRM has to be granted access, and Google Workspace and Microsoft Entra list every app someone has approved.
- Network and gateway logs, which show traffic to AI services nobody approved.
- Expense reports and API bills. A new AI subscription or API key shows up as spend.
- Your own domains. A new agent card or MCP server on your website is a public endpoint, and you can scan for it.
It isn't airtight. Personal accounts and keys people sign up for on their own won't show up in the admin console. So the best signal is still people coming to you. Make registering the easy path: register your app or agent and you get the approved connector and the shared key. Skip it and you're on your own.
Bring Security In as a Business Partner
If you're one of the new builders, don't wait for security to find you. Tell them what AI you're using, show them where your data goes, and bring them in early, the way you'd bring in finance or legal on a big decision. Security can't help with what it can't see.
Paste this into Claude, ChatGPT or whatever you use. It checks what you know about security, maps how your data flows, and drafts a note to your security team about the AI you're using. It doesn't replace that conversation. It gets you ready for it.
Your Website Is Part of the Surface Too
One more thing. AI agents don't only run inside your company. They're visiting your website, too. If your site publishes an A2A agent card or an MCP server, that's a public door into your systems. Treat it like any other app: review it, limit what it can do, and watch who uses it. You can run MachineReady to see what your site shows to AI, and send our A2A agent to your site to see what an agent finds there. For the bigger picture, I covered where AI agents should run and how to use ChatGPT for marketing securely.
Enablement and Security, Together
I don't want security teams to slow AI down. The business wins when more people can build. But the faster people build, the more the security team has to move from saying no to teaching the new builders, writing the rules down where the AI can read them, and checking the work before it goes out the door.
That's why AI is driving so much of the demand in cybersecurity right now. It isn't only the attackers using AI. It's every one of us building with it.
Want help writing AI security rules for your business? Get in touch. I have a few friends who can help you. And if you need help with your marketing, I'm happy to help.
*Sources: Verizon, "2026 Data Breach Investigations Report," executive summary, more than 22,000 confirmed breaches from October 2024 to November 2025. IBM and Ponemon Institute, "Cost of a Data Breach Report 2026," July 29, 2026, 602 organizations with breaches between March 2025 and February 2026. Gartner, "Gartner Forecasts the Market for Securing AI Will Reach $4.8 Billion in 2027," August 26, 2026. Edgile client figures from Wipro's acquisition announcement, December 20, 2021.*
← Back to all articles