Shadow AI incidents have more than doubled, while the audits that would catch them have declined.

Shadow AI has stopped being a hypothetical risk and started showing up in breach data

43% of businesses polled by IBM experienced a shadow AI cybersecurity incident in 2026. That’s more than double the previous year.

Shadow AI has transitioned from an abstract concern for IT governance into a countless number of attacks that actually happened, affected businesses worldwide, and led to measurable losses.

Take, for instance, Vercel’s supply chain data breach. Triggered by an unvetted AI-powered browser plugin, the incident led to a $2 million ransom. It’s just one of many real-life examples of shadow AI attacks that happened this year.

On average, such incidents cost businesses approximately $5.39 million each, up from $4.63 million the prior year. The consequences of these breaches vary:

• 49% resulted in data loss or compromise.
• 42% led to operational disruptions.
• 19% incurred regulatory fines.

These aftermaths are directly linked to three distinct exposures: a data problem (data loss), an availability problem (operational disruption), and a compliance problem (regulatory fines).

This complexity explains why shadow AI doesn’t fall neatly under a single business unit’s oversight, creating a tangled web of challenges for IT directors and business leaders alike.

Most IT Directors know that unsanctioned AI use is prevalent within their operations. However, they often can’t answer a more pressing question: “What AI tools are currently running in our environment?”

This article addresses that question. It explores the decline in shadow AI detection and provides insights on how to achieve full visibility across your business, transforming awareness into effective action.

If your detection efforts are going the wrong way, it doesn’t matter how much shadow AI there is in your business

Despite the alarming rise in incidents involving unsanctioned AI, the frequency of audits to monitor these activities has declined. According to the same IBM report, the percentage of businesses conducting regular shadow AI audits dropped from 34% to just 29%.

This indicates that nearly seven out of ten businesses haven’t implemented a systematic process to identify and track AI usage across their infrastructure and operations. This gap in auditing capabilities is critical, particularly as the incident rate has skyrocketed.

Why This Matters

A declining shadow AI audit rate doesn’t mean that businesses have grown less concerned. On the contrary, many are actively investing in employee awareness and written policies. But, at the same time, they are failing to investigate AI’s real-world utilisation within their business, missing significant dangers that could be hiding in plain sight.

In essence, businesses are spending valuable time and resources training employees about appropriate AI practices. Nevertheless, they are still falling short in understanding how their staff actually uses these technologies in practice.

IT Directors like you can’t control the prevalence of shadow AI, but can definitely increase your business’s detection capabilities.

AI adoption is harder to track than traditional IT. Here’s why.

AI adoption presents unique challenges compared to traditional IT systems, making it structurally harder to track. Unlike conventional IT, which can be systematically tracked, artificial intelligence (AI) tools often enter your business through unmonitored channels that make them more difficult to discover and control.

Here are some key factors contributing to this challenge.

Browser-based tools. Many AI tools work directly in web browsers, eliminating the need for installation or administrative rights. This ease of access lets employees circumvent established business protocols. For instance, your sales team might use an online AI tool to generate campaign emails faster. While this expedites the process, sensitive customer relationship management (CRM) data entered into these tools may remain on external servers, often unnoticed.

• Integrated AI features in SaaS platforms. Most Software as a Service (SaaS) applications now include built-in AI, which can be activated without oversight. When an employee enables these features on confidential documents, they unintentionally expose sensitive information. This integration generates unmonitored AI processes within already vetted systems.
• Personal devices and accounts. Employees might leverage their personal devices or accounts for work tasks using AI tools. One of your engineers could use his personal ChatGPT account for quick debugging, bypassing corporate monitoring. That’s what happened to Samsung in 2023.
• Browser extensions and agents. Employees often install browser plugins or agents without asking for permission. However, such software can interact with corporate systems. An agent designed to automate DevOps tasks can act on behalf of a user (e.g., an engineer) without supervision, potentially accessing databases or transmitting sensitive data to external services for processing. This lack of visibility puts your sensitive information at risk.
• Sessions rather than applications. Most AI interactions are session-based, leaving minimal residual data for tracking purposes. In the event of an incident, your IT team will have no way to verify what data was used, how it was processed, or why a decision was made.

Shadow IT vs. Shadow AI: Tangible vs. Intangible

Shadow IT is relatively easier to identify and spot because it’s usually tangible, such as the personal smartphones your employees use for work, or specific infrastructure. On the other hand, shadow AI is intangible and often characterised by session-based interactions that leave minimal trackable data.

For instance, an employee may use an external chatbot for technical queries instead of consulting a colleague or searching through your internal knowledge database. Since the chatbot is browser-based, once the user closes it, there’s nearly no record left, making it very unlikely that the activity will be flagged.

The Challenge of Organisational Oversight

To complicate things even further, AI capabilities frequently cross departmental boundaries. AI features are commonly bundled into several tools owned by different departments, such as IT, marketing, and finance. This fragmentation prevents any single department from gaining a holistic view of all AI tools in use.

Consider a scenario where your different departments have independently adopted AI-driven analytics tools. While your marketing team might leverage AI for customer engagement, the finance team utilises it for fraud detection.

Without centralised oversight, the tools’ overlapping functionalities will go unnoticed. This will lead to redundant subscriptions and potential compliance violations when handling sensitive data.

Ultimately, shadow AI discovery has become harder in precisely the way that makes deliberate instrumentation more necessary, not less. Consequently, understanding and taking action against shadow AI becomes more challenging.

Policy cannot govern tools that nobody is looking for

The rise of shadow AI presents significant challenges for businesses, leading many to implement acceptable-use policies as a primary response. Drafting a policy, communicating it, and training your workforce can be a reasonable initial step.

But these actions alone aren’t enough to mitigate the underlying risks associated with unapproved AI tools you don’t know exist.

3 Reasons Why Policies Alone Can’t Close the Oversight Gap

For effective policy enforcement and risk assessment, visibility into the AI tools used in your business is paramount. Policies alone fall short because:

1. Provide incomplete coverage. Policies are typically designed to govern known categories of tools and enforced through a control framework that covers only those applications already recognised by IT. So, if your employees use an IT-approved tool that comes with unapproved embedded AI features, the policy won’t cover it, leaving your business vulnerable to attacks.
2. Lack of awareness. Employees may use features in applications that are actually AI-driven without understanding their implications. For example, a worker may submit sensitive or proprietary data into a system without realising the feature he is using is AI-based and the data is processed or stored outside the business.
3. Don’t cover the oversight gap. Roughly 68% of UK businesses admit that employees regularly utilise unapproved AI tools. Yet, 72% of those businesses don’t monitor AI application usage in real time, and only about a third have a formal AI governance framework. Tools always outpace policy, and without a comprehensive view of AI applications used across your business, policy enforcement becomes useless.

Regulatory Implications

Under regulations such as the General Data Protection Regulation (GDPR) and the EU AI Act, businesses must demonstrate control over how data and AI interact. Undiscovered tools that escape oversight can’t be risk-assessed, nor can they be included in a Data Protection Impact Assessment (DPIA) or in incident reports.

This disconnect explains why 19% of shadow AI incidents have resulted in financial penalties. Businesses were fined not due to irresponsible AI use, but because they couldn’t produce documented proof of their compliance and oversight efforts.

This clearly demonstrates that simply having a policy in place isn’t enough to manage the threats posed by shadow AI.

Transform shadow AI discovery into a scheduled, instrumented process with a named owner

AI technology evolves faster than your standard yearly review. Vendors constantly add AI features to the SaaS tools your business relies on daily while your employees continuously seek out and access new AI-based applications.

That’s why shadow AI includes both the tools currently in use and those that will emerge in the future.

To manage this ecosystem effectively, you must therefore transform your AI discovery processes from occasional audits into a structured, ongoing effort. Mirror the methodology used in standard software vulnerability scanning. Assign a specific owner, create a detailed schedule, and set up a comprehensive remediation workflow.

AI Discovery Process by the Books: A Practical Example

Let’s say your business was the victim of a data breach because your HR department used unsanctioned AI software to analyse sensitive employees’ data.

If you had implemented a weekly monitoring schedule for AI tools and assigned a specific team member to oversee the process and update the IT team, you could probably have prevented the breach.

3 Key Elements for a Successful Discovery Process

Not all discoveries indicate a threat. Therefore, to build a successful discovery process that creates value:

1. Set a clearly defined cadence. Ensure your discovery process follows a precise schedule. Don’t initiate it only in response to incidents or when your board specifically asks for it.
2. Give owners authority to act. Empower the designated owner to make informed decisions about the tools discovered. It will foster responsibility for actions taken based on findings.
3. Implement a triage system. Create a review and sorting process for newly identified tools. Ensure the system allows endorsement of beneficial tools, sanctioning, blocking, or additional scrutiny for potential risks.

Building a Trustworthy Inventory

Implementing a punitive discovery process will drive your employees to hide their use of unsanctioned tools, resulting in costly non-compliance fines and operational inefficiencies.

Instead, foster an environment where your employees feel comfortable disclosing the AI tools they use. By emphasising a proactive, transparent, and continuously monitored discovery process, you will better manage shadow AI risks while promoting innovation and productivity.

Next steps: prioritise sequencing over scale

If, like many other IT Directors, you are facing budget constraints and limited headcount, manage shadow AI by prioritising sequencing over scale. Start by establishing a clear discovery baseline before updating your policies, purchasing new tools, or implementing new AI governance frameworks.

Each of these initiatives depends on an accurate inventory of AI tools and activities. So, an incomplete inventory can create persistent blind spots.

• Conduct a time-boxed discovery exercise. Use existing signals (e.g., network traffic data) to identify AI tools currently in use.
• Assign an owner. Designate a team member to oversee the discovery process and act on the findings.
• Set a regular cadence. Create a plan for repeated discovery efforts to ensure continuous monitoring.
• Create a decision-making route. Outline a clear process detailing what to do with discovered tools (e.g., sanctioning, blocking, or approving them).

Those businesses in the 29% conducting regular shadow AI audits aren’t necessarily more advanced. They are more successful in managing shadow AI simply because they follow a well thought schedule.

As shadow AI incidents skyrocket and audits decline, the most successful businesses will be those able to accurately answer: “What AI tools are running in our environment this month?”

If you struggle to answer that question, Acora’s data and AI governance expertise can help you establish a foundational baseline. Let’s build a new approach to AI visibility by replacing one-off audits with an ongoing discovery routine.