Using AI in companies: what is Shadow AI and how to act?

Half of the professionals who participated in the Panorama Vibe Coding Brasil survey, conducted by Skyone in partnership with the news in September 2026, had already created, on their own, an automation, application, dashboard, or integration without going through a formal IT project. Among these creations, 85.2% occurred without formal involvement from the technology area.
AI 16 min read By: Skyone

Half of the professionals who participated in the Panorama Vibe Coding Brasilsurvey, conducted by Skyone in partnership with the news in September 2026, had already created, on their own, an automation, application, dashboard, or integration without going through a formal IT project. Among these creations, 85.2% occurred without formal involvement from the technology area.

The numbers show a shift in how people solve problems within companies. With increasingly accessible artificial intelligence tools, professionals from different areas are able to develop solutions for their own needs, often even before involving IT.

The trend is also evident in how organizations guide this use. In the same survey, 42.9% of respondents stated that their companies do not have guidelines on what corporate data can be used in AI. Another 18.6% said that rules exist, but they are poorly communicated.

This scenario helps to understand a phenomenon that is gaining ground in discussions about corporate artificial intelligence: Shadow AI. When AI tools and solutions become part of the routine without adequate visibility, guidance, or supervision, the company loses the ability to monitor how the technology is being used.

This article explains how AI has reached teams, what characterizes Shadow AI, what risks can arise from uncoordinated adoption, and how to recognize existing use to establish guidelines.

How did AI become integrated into teams' routines without going through IT?

The path was short. Artificial intelligence tools allow professionals to solve specific problems without necessarily depending on a formal technology project.

An analyst who needs to organize information from a spreadsheet, a marketing professional who wants to automate a recurring task, or someone in sales who needs to develop a dashboard to track results can turn to AI to build a first version of the solution.

This behavior appears in the Panorama Vibe Coding Brasil survey. Half of the respondents have already created, on their own, an automation, application, dashboard, or integration without going through a formal IT project.

Among those who created these solutions, 51.1% worked individually, without involving IT, while 34.1% developed solutions with their own team, also without IT involvement. Only 14.8% reported receiving any occasional support from IT.

The results show that the ability to experiment with and build solutions using AI is becoming more widespread among different professionals and business areas.

The research also identified that this behavior is not restricted to a specific profile. Among the positions analyzed, self-creation appears in 62.7% of respondents who hold C-Level positions in IT, 57% among analysts and specialists, 55.1% among management and coordination professionals, and 50.5% among business C-Levels.

The change lies not only in the availability of the tools, but in the autonomy they offer. Professionals who previously had to present a need and wait for a solution to be developed can now test possibilities on their own.

This autonomy can accelerate processes and stimulate innovation. The challenge begins when solutions start circulating within the company without documentation, security criteria, or knowledge of the responsible areas.

What is Shadow AI?

Shadow AI is the use of artificial intelligence tools, models, or services within an organization without the knowledge, approval, or oversight of the areas responsible for IT, security, and data governance.

The concept encompasses different situations: an employee who enters corporate information into a public chatbot, a team that subscribes to an AI service without consulting IT, or a manager who uses an external tool to automate activities without knowing the data protection criteria involved.

What characterizes the phenomenon is the lack of visibility and coordination, not necessarily the tool used.

The News × Skyone research did not directly measure the occurrence of Shadow AI. However, its results help to contextualize the circumstances in which this type of behavior can emerge.

When 26.8% of respondents state that there is no official authorization for the use of AI and each person uses the tools on their own, without guidance, one of the challenges of corporate adoption becomes evident: the gap between what people do on a daily basis and what the organization is able to track.

Read also: Shadow IT 2.0: how to use AI with governance in companies

Shadow AI, Shadow IT, and BYOAI: what's the difference?

The three concepts are related, but they describe distinct situations.

  • Shadow IT is any technology used within an organization without the approval or knowledge of the IT department. It can involve cloud storage services, software, applications, and tools contracted directly by departments.
  • BYOAI (Bring Your Own AI) describes the behavior of bringing personal artificial intelligence tools into the workplace. It's a form of adoption that can happen without the company providing or approving a solution.
  • Shadow AI refers to the unauthorized or unsupervised use of artificial intelligence that may arise in this context.

The difference compared to traditional Shadow IT also lies in the type of information involved. In generative AI tools, the user can input text, spreadsheets, documents, and other corporate content to obtain answers or generate materials.

This means that the lack of visibility involves not only which tools are being used, but also what information flows through them and how the results are being used.

What are the risks of using AI without coordination?

The risks of indiscriminate adoption appear mainly on three fronts: data exposure, fragmentation of tools, and lack of visibility.

It is important to highlight that the Panorama Vibe Coding Brasil survey did not measure the occurrence of data leaks, security incidents, or financial losses associated with the use of AI. The following points represent possible risks given the conditions identified by the survey.

Data disclosure and intellectual property

When a person uses an AI tool without knowing the organization's security rules, they may end up sharing information that should remain restricted.

Imagine a professional uploading a spreadsheet with customer data to a public tool to request an analysis, or a colleague submitting an internal document summarizing its content.

Without defined criteria regarding what information can be used, the company loses some control over how this data is handled.

Research shows that this guidance remains a challenge. While 42.9% of respondents say there are no guidelines on what corporate data can be used in AI, only 26.3% say they have clear and formalized rules.

Another 18.6% report that the rules exist, but are poorly communicated.

The difference between having a policy and being able to apply it in daily practice deserves attention. Guidance that doesn't reach professionals will hardly be considered when they decide to use a tool.

Fragmentation of tools, accounts, and subscriptions

Without coordination, each team tends to choose the tools it considers most suitable for its own needs.

The result can be a set of overlapping signatures, personal accounts mixed with professional accounts, and differing guidelines on what can or cannot be shared.

Imagine, as a hypothetical example, three departments that contract different services to summarize documents. Each develops its own processes, accumulates a history of interactions, and works with different configurations.

Over time, the company may face difficulties in managing costs, standardizing usage, and maintaining the knowledge produced by the teams.

The same challenge arises when professionals develop automations, applications, and assistants without a common support structure.

This is one of the issues that enterprise AI platforms seek to address.

Skyone Astro, Skyone's corporate AI hub, brings together in a single environment access for teams to over 400 models, with configurable routing, as well as AI assistants adapted to the work context.

The proposal aims to allow each activity to use the most appropriate model, without depending on completely dispersed tools and access points.

Lack of visibility regarding who uses it, for what purpose, and with what data

One of the most important questions for leaders is also one of the most difficult to answer: who is using AI in the company, for what purposes, and with what information?

The Panorama Vibe Coding Brasil report shows that there is no single predominant model for guiding the use of technology.

Among the respondents, 25.1% say their companies officially encourage the use of AI and offer clear guidelines. Another 18.8% report that there is encouragement, but without defined rules. Meanwhile, 24.2% say that its use is permitted, but each area chooses its own tools, while 26.8% state that there is no official authorization and each person uses the tools on their own, without guidance.

The data show different levels of organization, but point to a common issue: the need to bring individual experimentation closer to corporate guidelines.

Answering questions about usage, purpose, and shared information requires monitoring mechanisms.

In Skyone Astro, the responsible area can view prompts and responses, apply filters, and export records in CSV format. The platform also offers data protection features, including blocking and alerts, as well as user-based data usage control.

This type of visibility helps to understand how AI is being used and to investigate situations that require attention. To function properly, however, it needs to be associated with clear rules and transparent communication with users.

Does banning the use of AI solve the problem?

In most cases, a blanket ban does not solve the issue of uncoordinated adoption.

Research by the news agency Skyone shows that experimentation is already part of the routine for many professionals. Half of the respondents created some kind of solution on their own, and 43% of those who have created solutions did so in the last six months.

The study also identified that 17.1% of respondents say that nothing prevents them from creating solutions and that they already do so when needed.

These results help to understand why simply restricting access to tools may not be enough. The need to solve problems still exists, and professionals may seek other paths when they cannot find suitable alternatives.

The challenge, therefore, is not to allow everything or block any initiative, but to establish conditions so that experimentation can happen responsibly.

This involves defining which tools can be used, what information is authorized for each activity, and in which situations IT involvement is necessary.

It is also important to differentiate between categories of data. Public information, internal documents, personal data, and trade secrets require different levels of protection and specific usage criteria.

The guidance needs to be clear enough for professionals to be able to apply it during their work, without relying on individual interpretations.

How do we recognize existing adoption practices and establish guidelines?

The starting point is recognizing that adoption is already happening. From there, five steps help transform scattered initiatives into an organized practice.

Start by listening to the teams

A brief internal survey helps identify which tools are being used, which tasks are performed with AI, and which solutions have already been developed.

The survey should be advisory in nature, not punitive. The goal is to understand the needs of the areas and identify opportunities to organize the use of technology.

Cross-reference the information with what appears in the systems

The teams' responses can be supplemented by analyzing expenses on corporate cards, reimbursements, browser extensions, and other available records.

In the case of internally developed solutions, it is also important to identify who created them, what process they support, what data is used, and who depends on that deliverable.

The goal is to build an inventory of the tools and initiatives that are already part of the operation.

You may also be interested in: Legacy systems and IT queues: what's holding back AI in companies?

Classify the use cases by data type and criticality

Summarizing a public document and analyzing a customer database are activities with different levels of risk.

Therefore, it is important to establish usage categories, considering the sensitivity of the information, the impact of potential errors, and the criticality of the process involved.

A simple classification, with activities that are unrestricted, permitted with care, and restricted, helps professionals understand which limits they must respect.

Publish clear guidelines and communicate with examples

A clear policy, with examples of permitted and prohibited situations, is usually more useful than a lengthy and difficult-to-consult document.

Research shows that 18.6% of respondents perceive that rules exist regarding the use of data in AI, but they are poorly communicated.

This reinforces the idea that simply publishing a policy isn't enough. People need to know where to find the guidelines, understand what they mean, and have a channel to ask questions.

Offer an approved alternative and designate someone to be responsible

Guidelines need to be accompanied by the right tools for the job.

If a company expects its teams to use AI responsibly, it needs to offer alternatives that meet day-to-day needs and allow for monitoring of usage.

It is also necessary to define who is responsible for the initiative, which areas participate in the decisions, and how the tools will be evaluated over time.

The topics of rules, permissions, and auditing are explored in greater depth in the second article of this series, on AI governance.

What would be a practical and successful alternative? An environment that caters to two audiences simultaneously.

For those who work with AI, it means having access to different AI models and assistants, choosing the most suitable resources for each task, and connecting tools used in their daily routine.

For administrators, this means setting limits, monitoring consumption, controlling permissions and connections, reviewing activities, and protecting information with blocking and alert mechanisms.

This is the Skyone Astro: to bring together autonomy for teams and governance for the business in the same environment.

However, the choice of any platform should stem from the organization's requirements and be evaluated in real-world usage situations, considering the needs of the teams and security criteria.

Conclusion: From spontaneous adoption to recognized adoption

Artificial intelligence is already part of how many Brazilian professionals solve problems at work. Data from the Panorama Vibe Coding Brasil survey shows that half of the respondents have already created solutions on their own, and that most of these initiatives occurred without formal IT involvement.

At the same time, the research reveals that 42.9% lack guidelines on what corporate data can be used in AI. This scenario indicates that the ability to experiment is advancing, while guidance does not always keep pace.

It is in this context that Shadow AI deserves attention. When the use of artificial intelligence tools and solutions occurs without visibility or shared criteria, the company finds it difficult to monitor what is being created, protect information, and transform individual initiatives into sustainable resources.

The answer doesn't have to be restricting experimentation, but rather recognizing what already exists, establishing guidelines, and providing conditions for teams to use technology responsibly.

Enterprise AI platforms, such as Skyone Astro, contribute to this process by bringing together access to models and assistants, auditing, data protection, and consumption control in a single environment.

Technology, however, is only part of the solution. The advancement of AI in companies also depends on guidance, training, and collaboration between departments.

Want to understand how to organize the use of AI in your company? Discover Skyone Astro and follow Skyone's content on governance and adoption of corporate AI.

Frequently Asked Questions

What is Shadow AI?

This refers to the use of artificial intelligence tools in a company without the knowledge, approval, or oversight of the areas responsible for IT, security, or data governance. Examples include AI assistants accessed with personal accounts, subscriptions contracted by departments without IT review, and browser extensions with AI capabilities.

Why are employees using AI without company approval?

The ease of access to tools and the ability to solve problems directly encourage spontaneous adoption. The Panorama Vibe Coding Brasil survey shows that 50% of respondents have already created an automation, application, dashboard, or integration on their own, without going through a formal IT project.

Does all use of Shadow AI cause data leaks?

No. The risk depends on the information used, the tool chosen, and the existing controls. Using it without guidance can increase the exposure of corporate information, but it doesn't mean that every initiative will result in a security incident.

Are Shadow AI and Vibe Coding the same thing?

No. Shadow AI is the use of artificial intelligence without the organization's knowledge or supervision. Vibe Coding is the practice of creating applications, automations, and other solutions with the aid of AI, using natural language to guide the construction. A Vibe Coding initiative can happen with or without IT involvement and does not necessarily constitute Shadow AI.

How can I find out if my company has Shadow AI?

Begin with an internal survey of the tools used, the activities performed, and the solutions developed by the teams. Cross-reference this information with expense records, corporate tools, and application inventories. It is also important to verify if there are clear guidelines on data usage and if professionals are aware of these guidelines.


Methodological note: the research was conducted with subscribers to the newspaper and gathered 1,971 complete responses. The results represent the survey participants and should not be generalized to all Brazilian professionals. Some cross-analyses use a base of 1,982 complete responses, as indicated in the study.1

Skyone
Written by Skyone

Start Your Digital Transformation Today

Transform Your Business with Skyone. Request a demo or schedule a call with our experts to discover how Skyone can accelerate your digital strategy.

Subscribe to our newsletter

Stay up to date with Skyone content

Contact Sales

Have a question? Talk to a specialist and get all your questions about the platform answered.