AI Breaches Cost $6 Million: 3 Security Markets Investors Should Watch

Higher breach costs give companies a reason to improve security. For investors, however, the useful question is which improvements customers will pay for and which suppliers can deliver them profitably. Answering that requires following the connection between a security weakness, a product, a purchasing decision and the financial results.

IBM Corp. (NYSE:IBM) provides the starting point. Its 2026 Cost of a Data Breach announcement reported that one in four malicious breaches involved AI-enabled techniques. Those incidents averaged $6 million, compared with $4.99 million across breaches overall.

These figures justify examining AI governance, identity controls and insider-risk monitoring. To evaluate those categories, begin with what the research actually measures, then work through the controls customers need, the budgets available and the evidence in company results.

1. Establish What The Breach Data Can Support

Start by distinguishing an attack that uses AI from an attack against an AI system. A criminal using a deepfake to mislead an employee creates a different security problem from someone manipulating a company’s AI assistant. The first may require stronger verification procedures; the second may require restrictions on what the assistant can access and execute.

This distinction matters because the approximately $1 million difference between IBM’s averages is not a measured governance penalty. The overall average includes different incident types, including AI-enabled incidents. It therefore does not provide a clean comparison between otherwise identical companies with and without AI controls.

The report examined breaches at 602 organizations between March 2025 and February 2026. That population helps describe losses among breached businesses, but cannot show how much any particular product reduces the chance of a breach.

Historical findings provide additional context. In IBM’s 2025 research, 63% of breached organizations lacked an AI governance policy or were developing one. Among those reporting compromised AI models or applications, 97% reported lacking AI access controls. These findings identify gaps within specific groups, rather than describe all companies or establish a verified 2026 trend.

Use the statistics to identify questions for further research. Do not convert them directly into sales forecasts or assumed savings. IBM sponsored the study and sells security products, making independent evidence of customer spending particularly useful.

2. Trace A Failure Before Choosing A Security Category

Once the evidence is clear, map the problem a customer needs to solve. A practical way to do this is to follow a single workflow and identify where an unsafe action could occur.

Consider a fictional assistant that reads supplier documents, retrieves customer information and sends email. A malicious document tells it to export confidential records. OWASP describes this mechanism as indirect prompt injection, where external content changes a model’s behavior.

The workflow becomes harder to examine when several agents divide the work. One retrieves files, another combines information and a third sends a message. Each action might look ordinary on its own, while the combined sequence creates a disclosure.

An investor reviewing a vendor should ask how its product would interrupt that sequence. Governance should establish the approved purpose and responsible owner. Access controls should limit permitted actions, while data controls should inspect transfers. Investigation tools should preserve enough information to explain what happened if prevention fails.

"AI is making attacks faster and cheaper, while breaches keep getting more expensive," said Suja Viswesan, vice president of IBM Security Software, in the report announcement.

The practical response is to examine the point where a product intervenes. Recording an incident, blocking an action and helping recovery are different capabilities. A vendor comparison should make those differences visible before comparing prices or market forecasts.

3. Evaluate What Each Product Actually Controls

AI Governance Should Connect Ownership With Action

Begin with whether the company can identify its AI systems and assign responsibility for them. The NIST AI Risk Management Framework offers a voluntary structure for managing these risks. Implementing such a framework creates a potential need for inventories, documented reviews and ongoing evidence collection.

IBM’s watsonx.governance documentation describes oversight and accountability capabilities. To assess the commercial value of those capabilities, look beyond the existence of a dashboard. Ask what happens when a connected application changes or an agent receives additional permissions.

Ben Hartwig, Web Operations Executive at Court Case Finder, explains: "Start with one AI workflow and ask who owns it, which information it can reach and who can stop it. Then change something in that workflow, such as adding a new data source. The useful test is whether that change triggers another review and reaches the people responsible for access. A policy document can describe the right process, but the company still needs evidence that the process happens when its systems change."

That test links the governance claim to a repeatable operational task. A platform that keeps inventories current and tracks unresolved exceptions has a stronger recurring-use argument than a tool used once to prepare a policy. However, recurring use becomes a revenue opportunity only if customers pay for it or renew because of it.

Check whether the supplier sells a separate subscription, includes the capability in a broader contract or relies heavily on consulting. Those arrangements can all create value, but they produce different revenue and delivery costs.

Identity Controls Should Limit Actions Throughout The Workflow

Governance establishes what should be allowed. Identity and access controls must then apply those decisions to the systems carrying out the work.

Authentication confirms an identity, while authorization determines its permitted actions. An agent with valid credentials can still cause damage if its permissions exceed its task. OWASP’s guidance on excessive agency identifies excessive functionality, permissions and autonomy as sources of risk.

Return to the supplier assistant. It may need to read invoices without being allowed to approve payments or export the customer database. Ask whether a vendor can enforce those limits in connected systems and withdraw access when the assignment ends.

Hartwig describes a useful evaluation sequence: "Do not stop at whether the agent can sign in. Check what it can read, what it can change and whether those permissions match the job it was given. Then test what happens when the owner leaves or the task ends. Access should not continue simply because nobody remembered to remove it. A useful review also follows the agent into connected systems, because a restriction in one application may leave another route open."

Okta Inc. (NASDAQ:OKTA) illustrates this category. Its September 2026 announcement described available agent identity capabilities alongside planned additions. Keep that distinction in the research notes: a roadmap explains direction, while availability, deployment and payment establish progressively stronger commercial evidence.

Benzinga’s coverage of secure identity infrastructure provides broader market context. For a specific investment, follow that context with a pricing question: does the contract charge for additional identities, usage or premium capabilities? Growth in agent numbers cannot establish revenue growth without that connection.

Data Controls Should Follow Information Beyond Login

Access restrictions reduce what an agent or employee can reach. They do not settle where information may go after an authorized person opens it. This is where data-loss prevention and insider-risk monitoring become relevant.

Microsoft Corp. (NASDAQ:MSFT) describes these functions in its Purview documentation for AI interactions. Its capabilities include controls for sensitive information and investigation of potentially malicious or inadvertent activity, with support varying by application and configuration.

For evaluation, separate prevention from investigation. A tool that blocks an inappropriate upload addresses the immediate transfer. A tool that connects several suspicious events may help investigators understand a wider pattern. Both can be useful, but an alert after disclosure should not be presented as preventing it.

Hartwig explains: "Follow a sensitive file through the whole task. Check whether the system recognizes the information, understands where it is going and can take an appropriate action before the transfer finishes. Then look at the investigation record. If a team cannot connect the user, the agent and the destination, it may struggle to understand the incident. Also test ordinary work, because a control that constantly blocks legitimate tasks can push people toward less visible workarounds."

Use that sequence when reading product documentation or customer case studies. Note which browsers, applications and automated connections are covered. Then examine investigation access, retention settings and alert workload, since these affect implementation cost as well as protection.

4. Find The Budget Behind The Security Need

After identifying a credible product, establish who would purchase it. Gartner Inc. (NYSE:IT) forecast in August 2026 that spending on securing AI would reach $4.78 billion in 2027, up 68.7% from approximately $2.84 billion in 2026. This supports investigating demand, but remains a forecast rather than booked vendor revenue.

"Traditional security tools often treat AI applications like any other software and need significant updates to address AI-specific threats," said Shailendra Upadhyay, senior principal analyst at Gartner, in the announcement.

The need for updates does not necessarily mean a larger total budget. IANS’ 2026 research, covering more than 500 security leaders, found median security budget growth of zero, while 69% identified AI as the leading priority for new investment. That priority also does not isolate protecting AI from using AI to improve defense.

IANS reported that 38% funded AI outside the security budget. A vendor may therefore need approval from technology or data leaders alongside security teams. Look for evidence that its sales process reaches the people controlling those funds.

Next, classify the purchase as additional spending, replacement spending or a bundled renewal. An illustrative $100,000 contract replacing $70,000 of older software raises the customer’s software budget by only $30,000 before implementation costs. The supplier wins revenue, but the industry does not gain $100,000 of net new demand.

This distinction adds detail to the broader discussion of AI investment and cybersecurity spending. It also explains why overlapping product categories should not be added together as independent market opportunities.

5. Check Whether Demand Reaches Financial Results

A credible security problem and an available budget establish a possibility. Company disclosures must show whether that possibility is becoming paid business.

Begin with customer evidence. Separate a free trial from a production deployment, then establish whether the deployment involves a paid upgrade or an existing entitlement. When those details are missing, record the uncertainty instead of treating a customer announcement as confirmed incremental revenue.

Next, compare the story with reported growth. In Okta’s quarter ended July 31, 2026, revenue rose 11% to $805 million. Subscription backlog expected to become revenue within 12 months rose 14% to $2.585 billion. These company-wide figures provide a baseline, but do not isolate AI-agent revenue.

That limitation should shape the next question: has management disclosed product-specific contributions, customer expansion or paid adoption that explains the AI connection? If not, avoid assigning the entire growth rate to a new product theme.

Then examine delivery economics. Runtime monitoring can involve computing, storage and specialist support. Ask whether pricing covers increasing usage and whether implementation becomes less labor-intensive as deployments expand. Revenue growth accompanied by heavy discounts or rising support requirements may have a weaker earnings impact than the headline suggests.

Finally, assess scale. A small product can grow rapidly without materially changing a large supplier’s results. Compare its disclosed contribution, where available, with total revenue and cash generation before judging its importance to the stock.

6. Define What Would Strengthen Or Weaken The Case

Turn the research into a short watchlist entry. State the control gap, the product addressing it, the likely budget owner, the available evidence of payment and the financial measure to revisit. This makes it easier to distinguish new evidence from repeated marketing claims.

A stronger case would show paid deployments expanding, customers renewing and delivery costs remaining manageable. A weaker case would show trials failing to convert, capabilities becoming bundled without additional payment or customers solving the problem through existing tools and procedures.

Review those signals alongside valuation. Even strong operating progress may disappoint investors if the share price assumes faster growth. The final judgment should connect the specific security need to observable business results and the expectations already reflected in the stock.

Disclosure: The author is independent and does not hold a financial interest, equity position, or professional relationship with IBM Corp., Okta Inc., Microsoft Corp., Gartner Inc., IANS Research, Court Case Finder, or any other companies mentioned in this analysis. This piece is written solely for analytical and educational purposes.

Benzinga Disclaimer: This article is from an unpaid external contributor. It does not represent Benzinga’s reporting and has not been edited for content or accuracy.