top of page
  • Linkedin

Marketing cybersecurity has become a business risk

10 minutes ago
18 min read
Infographic on marketing breach risks, linking marketing systems to customer data, CRM, finance and operations.
WSI - November 19, 2026

Why marketing cybersecurity matters for protecting customer data, advertising budgets and revenue reporting

Summary


Marketing has become an attractive cybersecurity target because of the customer information, advertising budgets and connected business systems it controls. As we introduce more AI tools and agents into everyday marketing operations, we're also creating new ways for sensitive information to be exposed, accounts to be compromised and automated systems to act with more authority than they should have.


This article looks at areas of exposure, including customer information being shared with AI tools, unreliable data entering advertising campaigns and AI agents with access to live accounts and publishing systems. It also examines how traditional cyberattacks are evolving with AI, why a problem that starts in marketing can affect other parts of the business, and what businesses and their marketing partners can do to reduce the risks without sacrificing the efficiencies these technologies provide.


Why it matters


  • Sensitive information is entering AI tools regularly. 

    • Cyberhaven's 2026 research found that employees were putting sensitive company information into AI tools approximately once every three days.


  • Employees aren't always clear about what can safely be shared.

    • Zapier's 2026 survey of 1,005 US workers at companies with more than 500 employees found that 79% had entered particularly sensitive information into AI tools, including customer or employee records, HR information and login credentials.


  • Unapproved AI use can increase the financial consequences of a breach. 

    • IBM's 2025 research found that one in five breached organizations experienced an incident involving unauthorized AI use, with high levels of that activity associated with an additional $670,000 in average breach costs.


  • AI access controls are frequently missing. 

    • Among organizations in IBM's research that experienced a breach involving an AI model or application, 97% lacked AI access controls.


  • AI agents introduce new opportunities for manipulation. 

    • OWASP ranks prompt injection, where malicious instructions attempt to influence an AI system, as the leading risk in its 2025 list of security vulnerabilities affecting large language model applications.


  • Mid-market businesses are significant targets. 

    • Black Kite found that mid-sized companies accounted for 73% of publicly disclosed ransomware and data-extortion incidents with known revenue in North America and Europe between January 2023 and June 2026.


  • Advertising accounts are already being targeted. 

    • Mimecast tracked 6.4 million detections over four years involving campaigns designed to steal Meta Business Manager and Google Ads accounts.


These findings come from different studies with different populations and methodologies, so they shouldn't be treated as directly comparable. IBM's research, in particular, examined organizations that had already experienced breaches. The figures describe risks observed within that population rather than the likelihood of a breach across all businesses.


AI is adding new ways in and marketing cybersecurity deserves more attention


We've spent years connecting marketing platforms so we can move information faster, manage campaigns more efficiently and understand which activities are actually producing revenue. Customer records flow between websites and CRM systems, advertising platforms use conversion data to make bidding decisions, and analytics tools bring everything together so we can see what's working. It's become such a normal part of running marketing that we don't always stop to think about how much sensitive information, financial control and access to other business systems now sits inside that environment.


And now we're introducing AI tools and agents into those same systems, sometimes giving them permission to retrieve customer information, analyze performance, adjust campaigns or publish content. The efficiencies are obvious, and I don't think anyone seriously expects businesses to stop adopting technology that can make them more productive. But I do wonder whether we've given enough thought to what happens when something goes wrong, particularly when an account is compromised or an AI application has access to information and systems it doesn't really need.


Cybersecurity has traditionally been treated as an IT responsibility, and understandably so. But marketing has become an attractive target in its own right, partly because of the information it holds and partly because of the money moving through advertising accounts. A stolen login, an unapproved AI tool or a compromised integration can create problems that extend well beyond a single campaign, especially when those systems are connected to customer databases, websites and financial reporting.


The risks aren't all the same, and I think that's an important distinction. Some involve unauthorized access or stolen information, others involve fraudulent activity that distorts campaign performance, and still others arise because automated systems have been given more authority than they need. What connects them is our growing dependence on marketing technology that controls valuable information and makes decisions with real financial consequences.


When we think about what an attacker might want from a business, customer databases and financial information probably come to mind first. But consider what someone could potentially reach through a marketing account. Depending on how the business is set up, there may be customer lists, CRM records, advertising audiences, pricing information, website publishing permissions, analytics data and access to live advertising budgets. Agencies, employees, contractors and software providers may all have some level of access, and over time those permissions can accumulate as people change roles, suppliers come and go, and new applications are connected.


None of that automatically makes the environment insecure. Businesses need these connections to operate efficiently, and most are there for perfectly legitimate reasons. The concern is whether anyone has a complete picture of what's connected, who can get into it and how far a compromised account could reach. An attacker who gains access to an advertising account doesn't necessarily gain access to the CRM or the company's financial systems, but if credentials are shared or integrations have broad permissions, the exposure can become much larger than the original point of entry.


There's evidence that attackers understand the value of these accounts. Mimecast tracked 6.4 million detections over four years involving campaigns designed to steal Meta Business Manager and Google Ads accounts, while Bitdefender documented cases where hijacked Google advertiser accounts were used to distribute malicious advertisements. These are particularly attractive targets because they provide access to established advertising infrastructure and, potentially, budgets that can be redirected toward fraudulent activity.


For businesses, the damage isn't necessarily limited to the money spent before someone notices. Campaigns may need to be suspended, access credentials replaced and account activity investigated, while the marketing team works out whether conversion tracking, customer information or other connected systems have also been affected. Even when the breach is contained, the disruption can be expensive, especially for companies that depend on a steady flow of leads or online sales.


Customer information is finding its way into AI tools


I suspect one of the most common exposures doesn't look particularly risky when it happens. Someone in marketing needs to organize a customer list, compare campaign results or pull together a presentation, and rather than spend an hour working through the information manually, they paste it into an AI tool. It saves time, the output looks good and the work gets done. The problem is that nobody may have considered whether the information was appropriate to share, whether the tool was approved or what happens to the data once it's been entered.


This is becoming a much more common issue as AI moves into everyday business operations. Cyberhaven's 2026 research found that employees were entering sensitive company information into AI tools approximately once every three days. Zapier surveyed 1,005 US workers at companies with more than 500 employees and found that 79% had entered particularly sensitive information into AI, including customer or employee information, HR records and login credentials. The studies have different populations and methodologies, so I wouldn't combine their findings into a single statistic, but they do suggest that the use of AI is moving faster than many organizations' ability to manage it.


What interests me is why employees are doing this in the first place. I don't believe most people are deliberately putting their employers at risk. They're trying to get work done, and if an AI tool makes a tedious task easier, they're going to be tempted to use it. Zapier's research found that half of the workers surveyed had ignored workplace AI restrictions, and among those using personal AI accounts for work, many preferred the experience of their own tools or found the approved alternatives too difficult or slow. That tells me this isn't simply a matter of writing a policy and expecting everyone to follow it.


Businesses need to make approved tools practical enough that people actually want to use them, while also being very clear about what information can and can't be shared. Customer records, credentials, regulated information and commercially sensitive material need appropriate safeguards, and employees need training that reflects the work they're doing rather than a generic annual reminder about cybersecurity. Zapier found that employees who felt confident about what was safe to share were much more likely to have received formal training, which suggests that education has a meaningful role alongside technical controls.


The other problem is that having a policy doesn't necessarily mean anyone is checking whether it's working. IBM's 2025 Cost of a Data Breach research found that among breached organizations with AI governance policies, only 34% regularly audited for unauthorized AI use. That figure comes from a study of organizations that had experienced breaches, so it shouldn't be interpreted as a measure of all businesses, but it does highlight a weakness worth considering. It's difficult to protect information when we don't have enough visibility into where it's going.


The information coming into marketing can be just as problematic


Most cybersecurity discussions concentrate on preventing information from being stolen or exposed, but there's another issue in marketing that deserves attention, even though it isn't necessarily a cybersecurity breach. It's the integrity of the information coming into our advertising platforms and the decisions those platforms make based on it.


Think about how much campaign optimization now depends on automated bidding. We tell Google or Meta what outcome we're looking for, whether that's a purchase, a qualified lead or another conversion, and their systems use incoming signals to determine where advertising budgets should go. When those signals represent genuine customer activity, the technology can be extremely effective. But when bots, click farms or automated scripts generate invalid clicks, form submissions and conversion events, the information feeding those decisions may not reflect real demand.


I've written about invalid traffic separately because I think it's one of the more overlooked issues in paid media. A campaign can appear to be performing well, with conversion numbers moving in the right direction, while the sales pipeline tells a very different story. Some of that discrepancy may be ordinary measurement error or poor lead quality, but invalid activity can also contribute, particularly when it isn't being identified and filtered effectively.


It's important not to confuse this with someone breaking into an advertising account. Invalid traffic is primarily an advertising fraud and data integrity problem, and it can happen without any unauthorized account access. But the financial consequences overlap with the broader security discussion because we're relying on automated systems to make spending decisions, and those decisions are only as good as the information they receive.


The more we automate campaign management, the more important it becomes to validate what the platforms are reporting against actual customer activity and revenue. A conversion event is useful, but it isn't necessarily a customer, and a dashboard full of positive numbers doesn't mean much if the business isn't seeing the corresponding results.


AI agents introduce another level of access


There's a considerable difference between using an AI tool to help write a campaign report and giving an AI agent permission to work inside a live advertising account. In the first situation, we're generally asking the technology to produce information that a person can review. In the second, we're potentially allowing it to take action, whether that's changing a budget, updating a campaign, publishing content or retrieving information from connected systems.


I can see why businesses are interested in these capabilities. There's enormous potential to reduce repetitive work and make marketing operations more responsive, particularly for smaller teams that are already stretched. But the conversation about what an agent can do needs to include what it should be allowed to do, and what protections exist if the information it encounters is misleading or deliberately malicious.


One of the risks receiving considerable attention is prompt injection, which is essentially an attempt to manipulate an AI system through instructions hidden in material it reads. An agent might be reviewing a web page, processing an email or examining customer feedback when it encounters instructions designed to make it behave differently from what its operator intended. OWASP has identified prompt injection as a leading risk for large language model applications, and the concern becomes more serious when an agent has permission to make changes or access sensitive information.


This isn't quite the same as traditional phishing, although the comparison is useful. Instead of persuading a person to click a malicious link or disclose a password, an attacker is trying to influence an automated system. The difficulty is that agents often need to read information from outside sources to do their jobs, and we can't assume everything they encounter is trustworthy.


That's why I think permissions matter more than many of the conversations we're having about AI productivity. An agent that analyzes campaign performance doesn't necessarily need permission to change budgets, and one that reads customer feedback shouldn't automatically be able to publish content or export customer records. Separating those responsibilities may add a little friction, but it also limits what can happen if the agent is manipulated or compromised.


There are also the credentials these systems use to connect with other platforms. AI agents and applications often rely on access tokens, which function as digital credentials for advertising accounts, CRM systems, websites and other services. If those tokens are exposed through insecure logs, shared documents or another vulnerability, someone who obtains them may be able to use the permissions attached to them.


I've discussed AI agent accountability in an earlier article, particularly the question of who should authorize automated decisions. The issue here is more immediate. Before connecting an agent to business systems, we need to understand what information it can reach, what changes it can make and how that access is protected.


The traditional attacks haven't gone away


With so much attention on AI, it's easy to overlook the fact that many of the most serious cybersecurity incidents still begin with familiar problems. Phishing, stolen credentials, outdated software and poorly managed third-party access haven't disappeared, and attackers are now using AI to make some of those techniques more convincing.


KnowBe4 reported in April 2026 that 86% of the phishing campaigns it examined over a six-month period involved some use of AI. That doesn't mean every attack was created entirely by AI or that the technology caused every compromise, but it does suggest that attackers are incorporating it into methods businesses have been dealing with for years.


Advertising accounts are particularly appealing because they can be exploited quickly. An attacker who gains control of an established advertiser account may be able to launch fraudulent campaigns, redirect spending or use the account's existing credibility to distribute malicious advertisements. The financial exposure is immediate, and the business may not discover the problem until unusual activity appears in its account or billing records.


Multi-factor authentication remains an important safeguard, but it shouldn't be treated as an absolute guarantee. Attackers have developed techniques that can intercept authentication information or bypass certain implementations, which is why strong authentication needs to sit alongside access reviews, secure credential management and monitoring for suspicious activity.


The same attention needs to extend to the software and suppliers connected to marketing. Verizon's 2026 Data Breach Investigations Report found that vulnerability exploitation accounted for 31% of initial access methods, while third parties were involved in 48% of breaches. Those are broader cybersecurity findings rather than marketing-specific statistics, but they're relevant when we consider how many external applications, website plugins, tracking tools and service providers have some level of access to marketing infrastructure.


An outdated WordPress plugin or a former agency login may not seem particularly significant during a busy marketing week, but both can create unnecessary exposure if they're neglected. And because marketing systems are often connected to customer databases, analytics platforms and other business applications, the potential consequences depend on what those connections allow someone to reach.


A breach that starts in marketing may not stay there


This is probably the part of the discussion that concerns me most, because marketing has become so interconnected with the rest of the business that we don't always recognize where its boundaries are anymore.


Advertising platforms connect to websites and analytics, CRM systems connect to customer databases and email platforms, and conversion tracking often feeds information into reporting systems used by sales, finance and management. We've built these connections for good reasons. They help us understand customer behaviour, improve campaign performance and make better decisions about where to invest. But every integration also raises questions about access, permissions and what could happen if one of those connections were compromised.


A breach in marketing doesn't automatically mean an attacker can move into finance, operations or every other connected system. That depends on how the technology is configured, whether credentials are shared and what permissions have been granted. Still, I think businesses need to understand those pathways rather than assume that a marketing account is isolated simply because it belongs to a different department.


The consequences can also be indirect. An incident might not give an attacker access to financial systems, but it could still disrupt advertising activity, expose customer information or interfere with the tracking data management uses to assess revenue performance. If conversion events are altered or reporting configurations are changed without authorization, the business may need to investigate the accuracy of its numbers before relying on them again.


For a mid-market company, that kind of disruption can be difficult to absorb. Advertising campaigns may need to stop while access is secured, employees may be pulled away from revenue-generating work to investigate the problem, and customer confidence can suffer if sensitive information is exposed.


Black Kite's 2026 research found that mid-sized companies accounted for 73% of publicly disclosed ransomware and data-extortion incidents with known revenue in North America and Europe during the period it examined. That doesn't mean 73% of mid-market companies experienced an attack, but it does challenge the idea that cybersecurity is primarily a concern for large enterprises with extensive IT infrastructure.


What good marketing security should look like in practice


I don't think most businesses need another complicated framework that looks impressive in a presentation but never makes it into day-to-day operations. What they need is a clear understanding of where their information lives, who has access to it and what safeguards exist around the systems that control customer data, advertising budgets and revenue measurement.


Much of that comes down to a handful of practical disciplines that should already be part of responsible marketing management.


  1. Know who and what has access 

    • Employees, agencies, contractors, software integrations and AI agents should have permissions that reflect what they actually need to do. Those permissions should be reviewed regularly, particularly when people change roles, suppliers are replaced or applications are no longer being used.

  2. Make approved AI tools practical to use 

    • It's difficult to enforce policies when the approved alternatives make everyday work unnecessarily complicated. Businesses need clear guidance about sensitive information, training that reflects real working situations and tools that employees can use without looking for workarounds.

  3. Separate information access from decision-making authority 

    • An AI agent may need to review advertising performance without having permission to change budgets or publish content. The same principle applies to people and external partners. Access should be appropriate to the task rather than granted broadly for convenience.

  4. Protect the accounts that control money and customer information

    • Multi-factor authentication, secure credentials, software updates and regular access reviews should be standard across advertising platforms, websites, CRM systems, analytics and email. Old accounts and unnecessary permissions need to be removed rather than left in place indefinitely.

  5. Keep ownership of marketing data with the business 

    • Advertising accounts, analytics properties and conversion tracking should be structured so the business retains control of its information and can manage access independently of any particular agency or supplier. GA4, for example, should sit within a business-controlled account, with appropriate permissions for everyone who needs to work with it.

  6. Validate the information driving advertising decisions

    • Conversion tracking, form submissions and campaign performance should be checked against genuine lead quality, customer activity and revenue. Invalid traffic protections and regular measurement reviews help reduce the risk of automated bidding systems learning from misleading signals.


None of these measures guarantees that an incident won't happen, and I wouldn't suggest otherwise. But they make it easier to understand where exposure exists, reduce unnecessary access and identify problems before they become more serious.


Agencies and technology partners are part of the security picture


There's another aspect of this that I think deserves more attention, particularly because so much marketing work is now delivered through a combination of internal teams, outside agencies and specialist technology providers.


Agencies often need privileged access to advertising accounts, websites, analytics platforms and customer information to do their jobs properly. That's entirely reasonable, but it also means security can't be treated as something that belongs exclusively to the client's IT department. The way we manage credentials, assign permissions, introduce new applications and remove access when an engagement ends has a direct bearing on the client's exposure.


The same applies when agencies begin using AI tools or agents as part of campaign management, reporting or content production. Businesses should know which systems are being used, whether customer information is involved and what authority those applications have. They shouldn't have to discover after an incident that an external tool had access to information or accounts they didn't realize were connected.


I also think we need to be more transparent about measurement. If we're using automated bidding systems to manage advertising investment, we should be able to explain how conversion data is collected, what protections exist against invalid activity and how reported performance connects to actual business results. That isn't just a reporting issue. It's part of protecting the integrity of the information used to make financial decisions.


These are reasonable conversations for businesses and their marketing partners to have together, and they're much easier to address before something goes wrong than after everyone is trying to work out who had access to what.


The security conversation needs to catch up with marketing


We've made enormous progress in connecting marketing technology, and I wouldn't want to go back to the days when every platform operated independently and reporting required hours of manual work. Those connections have helped businesses become more efficient, understand their customers better and make more informed decisions about advertising investment.


But we've also created an environment where marketing controls valuable information, significant budgets and access to systems that extend into other parts of the business. AI is adding another layer of capability, particularly as agents begin doing work that previously required someone to log in, review information and make a decision themselves.


I think that's where the conversation needs to change. We can't keep treating marketing security as a collection of passwords and software updates while simultaneously giving more people, platforms and automated applications access to the systems behind customer acquisition and revenue reporting.


This isn't an argument against AI or connected marketing technology. It's an argument for understanding what we've built, how the different parts interact and where the risks sit before we're dealing with the consequences of an incident.


Because if someone gained unauthorized access to one of our marketing systems tomorrow, I'd want to know not only how quickly we'd detect it, but whether we could confidently explain what else they might be able to reach.


And I wonder how many businesses could answer that question today.



Frequently asked questions about marketing cybersecurity



What are the biggest cybersecurity risks in digital marketing?

The biggest cybersecurity risks in digital marketing include stolen advertising account credentials, phishing attacks, customer data exposure, vulnerable website software and unauthorized access through third-party tools. AI is introducing additional concerns as employees share information with external applications and automated agents gain permission to access or change marketing systems. Because advertising platforms, websites, CRM systems and analytics are often connected, a compromised account can potentially affect more than one part of the business. The extent of that exposure depends on how the systems are configured and what permissions have been granted.

It depends on the information being shared, the AI platform, its privacy settings and the safeguards the business has in place. Customer names, contact information, purchasing histories and other personal data shouldn't be entered into an AI tool without confirming that the use is authorized and appropriately protected. Some AI services offer business-level data controls, but those protections aren't necessarily the same across free, personal and enterprise accounts. Before using AI with customer information, businesses should review the provider's data retention policies, training settings, access controls and contractual protections. Where possible, sensitive information should be removed or anonymized before it's shared.

Start with strong authentication, individual user accounts and regular reviews of who has access. Shared passwords should be avoided, former employees and agencies should have their permissions removed, and third-party applications should only receive the access they actually need. It's also worth monitoring campaign changes, billing activity, new users and connected manager accounts for anything unusual. If an account is compromised, the business should secure its credentials, investigate unauthorized activity and report the incident to the advertising platform as quickly as possible. These precautions help reduce the risk of account hijacking, fraudulent advertising spend and disruption to active campaigns.

Prompt injection is a security risk where malicious instructions are hidden in information an AI system reads, such as a website, email, document or customer review. The attacker attempts to persuade the AI to behave differently from its original instructions. In marketing, this becomes particularly concerning when an AI agent has permission to access customer records, change advertising campaigns or publish website content. A successful attack could potentially cause the agent to disclose information or take an unauthorized action. Businesses can reduce the exposure by limiting agent permissions, separating data access from decision-making authority and requiring human approval for sensitive actions.

Yes. Bots, click farms and automated scripts can generate invalid clicks, form submissions and conversion events that make campaign activity appear more valuable than it really is. When invalid conversions aren't identified and filtered, automated bidding systems may receive misleading performance signals, potentially affecting how advertising budgets are allocated. This is primarily an advertising fraud and data integrity issue rather than necessarily a cybersecurity breach. Businesses should compare platform conversions with qualified leads, actual customer activity and revenue, while reviewing invalid traffic protections and the accuracy of their tracking configuration.

A breach that begins in marketing can potentially affect other business systems when accounts, credentials and applications are connected. For example, compromised access to a marketing platform might expose customer records, disrupt website activity or allow unauthorized changes to campaign tracking and reporting. It doesn't automatically mean an attacker can reach finance, operations or every connected application, because that depends on the permissions and security controls in place. However, even a contained marketing incident can lead to wasted advertising spend, interrupted lead generation, unreliable reporting and damage to customer trust. Understanding how marketing systems connect to the wider business is an important part of managing that risk


Sources

  • IBM, Cost of a Data Breach Report 2025, July 30, 2025

  • Cyberhaven, Sensitive Enterprise Data Is Flowing Into AI Tools at Scale, February 11, 2026

  • Zapier, 79% of Workers Have Given AI Sensitive Info, July 29, 2026

  • Nationwide research, reported by Under30CEO, Small Business Cybersecurity Lags Your Team's AI Habit, September 17, 2026

  • OWASP GenAI Security Project, Top 10 for Large Language Model Applications, 2025

  • KnowBe4, Prompt Injection and the Rise of Agentic Risk, July 9, 2026

  • Verizon, 2026 Data Breach Investigations Report

  • Black Kite, Mid-Market Is the Routine Target, August 18, 2026

  • KnowBe4, Phishing Threat Trends Report, April 30, 2026

  • Mimecast, Ad Account Theft: The Digital Advertising Commodity Fueling Global Fraud, July 28, 2026

  • Bitdefender Labs, Hijacked Google Ads Push Fake Software Downloads to Mac Users, February 18, 2026

  • BleepingComputer, Fake ChatGPT, Gemini Sites Steal Advertising Accounts and MFA Codes, October 6, 2026

Comments


bottom of page