- Privacy and artificial intelligence: why has it become an urgent issue for SMEs?
- How widespread is AI in Italian companies, and how well is it actually governed?
- How does data privacy really work in a company?
- What really changes when you introduce AI in your company?
- What does the AI Act require of Italian SMEs?
- What are the most common privacy and artificial intelligence risks for SMEs and professionals?
- How much do GDPR violations really cost in Italy?
- What can SMEs do in practice to protect data without blocking AI?
- How do you define internal policies on AI use?
- How do you separate sensitive data from public AI tools?
- How do you vet AI software and vendors before adopting them?
- Three real-world scenarios: how do the risks change for an e-commerce store, a professional firm and a manufacturing company?
- Why has company data become a strategic asset, and why does protecting it pay off?
- How do you get started: the steps to put privacy and AI in order in your company
- What mistakes are made most often when tackling this topic?
- How do I know if my privacy and AI governance is working?
- So, is it worth investing in privacy and AI governance for your company?
- How we see it at SHM
- Frequently Asked Questions
- AI adoption among Italian SMEs has doubled in one year (from 7.7% to 15.7%), but 76% have not yet invested or plan to: the gap between those who move and those who wait is widening.
- The main risk is not an external cyberattack: it is employees unknowingly uploading sensitive data to public tools (Shadow AI).
- Defining internal policies, mapping data flows and training staff reduces most risks without blocking the use of AI.
- GDPR fines in Italy grew by 54.5% in one year: ignoring data governance has an increasingly real cost.
Updated on November 9, 2026
34% of Italian workers use AI tools their company hasn't approved, and they often do so even when the company already provides one. The problem isn't the technology: it's the lack of clear rules on what can be uploaded and where.
Privacy and artificial intelligence: why has it become an urgent issue for SMEs?
AI has entered daily workflows much faster than companies have had time to regulate it. Every prompt sent to an AI tool is an exchange of information: without clear rules, that data can end up where it shouldn't.
AI adoption in SMEs is growing much faster than companies' ability to govern their data management processes and criteria. Generative tools, AI assistants built into business software, smart automations and analytics platforms are entering daily workflows without requiring advanced technical skills. That makes AI accessible even to small and mid-sized businesses, but it also raises the risk of poorly controlled use.
Every interaction with an AI system creates an exchange of information: a prompt can contain commercial data, internal documents, customer emails, technical specs or financial information. In many companies, this content is uploaded without checking where it's processed, how long it stays stored or which integrations are active between different platforms.
More than regulatory compliance in the strict sense, the issue is control: protecting know-how, keeping your digital infrastructure in order and knowing who accesses what. SHM Studio supports SMEs and professionals in building AI ecosystems that are more controlled, sustainable and aligned with business goals.
| Your situation | Urgency | why |
|---|---|---|
| You use public AI tools without internal policies | No, for now | Company data may end up in third-party archives with no control |
| You have team members who use personal AI accounts for work | No, for now | Shadow AI: no traceability, no protection |
| You have mapped the data flows between your platforms | Yes | You know where the risk applies and where it doesn't |
| You have written policies on which data can be uploaded to AI | Yes | Reduces improvised behavior and legal risks |
| You use AI solutions in closed or private environments | Yes | Greater control over data, access and information flows |
How widespread is AI in Italian companies, and how well is it actually governed?
Adoption has doubled in a year, but governance hasn't moved. Most Italian SMEs use AI with no structured training and no defined policies.
According to the Istat survey "Imprese e ICT – Anno 2025", AI adoption among Italian businesses with at least 10 employees rose from 5.0% in 2023 to 8.2% in 2024, and then to 16.4% in 2025: practically a doubling in a single year. The figure for SMEs is even starker: from 7.7% to 15.7% in twelve months.
Yet according to the 2025-2026 Research by the Osservatorio Innovazione Digitale nelle PMI at Politecnico di Milano, 76% of Italian SMEs have neither invested nor plan to invest in AI, and only 7% have launched structured training programs. This means those who do adopt AI often do so without an organizational structure able to support it.
For many SMEs, privacy and artificial intelligence is therefore a problem that already exists, not a future one. To dig deeper into the practical adoption of these tools, our guide to artificial intelligence for SMEs offers an up-to-date overview of the most useful tools in 2026.
How does data privacy really work in a company?
Business privacy isn't just about cookie policies and consent: it covers every flow of information across email, CRM, cloud and AI tools. Without a clear map of these flows, you can't know where your weak points are.
When we talk about business privacy, legal documents and cookie policies come to mind first. But more everyday aspects matter too: customer information, sales databases, contracts, financial data, login credentials and proprietary content move constantly between email, CRM, cloud platforms, management systems and collaboration software. Each of these is a potential point of access, alteration or loss of information.
The GDPR (the European regulation on personal data protection) clearly defines who can access data, where it's stored, how it's processed and which tools handle it. That's why privacy and artificial intelligence are becoming closely linked: AI tools work on the data they receive, and without a clear organizational structure, the chances of misuse or poorly controlled information flows go up.
The more software you connect, the greater the need to govern access, data flows, permissions and internal processes. It's not just a big-company problem: SMEs are often more exposed precisely because they have fewer resources dedicated to oversight.

What really changes when you introduce AI in your company?
AI isn't just another piece of software: it changes how information is processed. Data entered in a prompt can become part of the technology provider's archive, with consequences many companies haven't yet considered.
Introducing artificial intelligence changes the relationship between a company and its data because it transforms how information is processed. Many AI tools read the content entered in prompts to train their models or to give accurate answers. In the process, the data a user enters can become an integral part of the technology provider's archive.
There is an important technical difference between two categories of tools:
- Public tools: designed to learn from user interactions. The data you enter may be used to improve the model.
- Enterprise solutions in closed environments: configured not to use incoming information for training. They offer greater control over data.
One critical issue is Shadow AI: employees, trying to speed up their work, use personal AI software accounts without authorization or oversight. According to the HR Innovation Practice Observatory at Politecnico di Milano, the share of Italian workers using AI tools not provided by their company rose from 27% to 34% in a year. The most surprising finding: 85% do so with personal or free tools even when the company already provides one.
Without internal rules, information ends up in third-party databases with no protection at all. To see how this ties into broader security risks, our analysis of autonomous AI models and compliance risks offers a useful overview.
What does the AI Act require of Italian SMEs?
The AI Act, the European Union's first comprehensive regulation on artificial intelligence, classifies applications by risk and sets clear obligations for transparency, documentation and human oversight. For SMEs, this means building better processes before even choosing the tools.
The AI Act requires companies to manage their systems with a risk-based approach. Applications are classified by their potential impact on citizens' fundamental rights and safety. Software providers are now required to ensure transparency in how their algorithms work, documenting the datasets used for training and putting human oversight measures in place.
If your company integrates solutions classified as high-risk, such as those used for hiring, credit scoring or managing critical infrastructure, the regulation requires a quality management system and accurate logging to ensure full traceability of decision-making processes.
Technical documentation and data management are therefore a must before integrating any AI-based solution into your IT systems. Ignoring this exposes you to significant fines and unwanted disruptions to your workflow.
| AI Act risk category | Application examples | Main obligations |
|---|---|---|
| Unacceptable risk | Mass biometric recognition, behavioral manipulation | Prohibited |
| High risk | Recruitment, credit scoring, critical infrastructure | Documentation, logs, human oversight, quality system |
| Limited risk | Chatbots, content generators | Transparency obligation toward the user |
| Minimal risk | Spam filters, AI in video games | No additional specific obligations |
What are the most common privacy and artificial intelligence risks for SMEs and professionals?
Most issues don't come from sophisticated cyberattacks, but from everyday behaviors with no clear procedures: accidental data uploads, uncontrolled access, unverified software.
Before introducing new platforms, companies need to understand where their digital processes are most vulnerable. The most common risks fall into four main categories.
- 1 · Employees and teamsUnintentional sharing of sensitive dataSales emails, contracts, price lists and customer data get uploaded to public AI tools to speed up everyday tasks, without considering where that content is processed and stored.
- 2 · Management and ITUncontrolled access to AI toolsShared accounts, informally managed credentials, external contractors using company tools without supervision: without governance, it's impossible to monitor who uses AI systems and what data they process.
- 3 · Individual usersUse of unvetted softwareBrowser extensions, email plugins or free tools of dubious origin often ask for permission to read everything on your screen, turning into an open door to external databases.
- 4 · The company as a wholeReputation and compliance issuesUncontrolled data handling can damage customer trust and company reputation, especially in sectors where information processing is a core part of the service.
Shadow AI deserves a closer look. According to Politecnico di Milano, 51% of Italian workers supplement or replace company solutions with external tools. This makes any preventive confidentiality control essentially ineffective unless it comes with a shared organizational culture.
For those running digital campaigns, the risk extends to advertising data: contact lists, audience segments and CRM data uploaded to unverified platforms can violate both the GDPR and the platforms' own policies.
How much do GDPR violations really cost in Italy?
Fines for privacy violations in Italy grew by 54.5% in one year. It's not just large companies: SMEs are on the regulators' radar too.
In 2025, the Italian Data Protection Authority (Garante per la protezione dei dati personali) adopted 807 collegial decisions and collected fines totaling 37.7 million euros, up 54.5% from 24.4 million in 2024. Data breach notifications reached 2,415, a 10% increase over the previous year.
Across Europe, according to Federprivacy, European authorities issued 335 fines for GDPR violations in 2025, totaling over 1.1 billion euros. Excluding the 10 largest fines, the average drops to 298,000 euros: a much more realistic amount for mid-sized businesses, but still significant.
The most relevant figure for SMEs isn't the size of individual fines, but their frequency: 506 corrective and sanctioning measures in a single year means authorities are stepping up checks across all sectors. The issue of data retention in AI tools is one of the areas where regulators are focusing most closely.
What can SMEs do in practice to protect data without blocking AI?
Addressing the relationship between privacy and artificial intelligence doesn't mean blocking new technologies. It means setting clear rules on who can use what, with which data and in which environments.
SMEs can reduce most risks by working on internal organization and access management. The starting point is understanding how data actually moves through your company: customer information, sales documents, databases and operational content constantly pass between different platforms. Without a clear map of these flows, it's hard to know where to act.
Privacy and artificial intelligence should be treated as a technology governance issue: that means defining policies, permissions, review procedures and usage criteria shared across management, staff and external vendors.
| Area of intervention | Concrete action | Priority |
|---|---|---|
| Internal policies | Define which tools are authorized, which data can be uploaded, and who can access what | High |
| Data separation | Keep sensitive data out of public tools; anonymize data beforehand | High |
| Vendor vetting | Check policies, server location, data retention and whether your data is used for training | High |
| Access management | Define roles, permissions and authorization levels for each AI tool | Medium |
| Staff training | Give simple, practical guidance on what people can and can't do with AI tools | Medium |
| Private AI environments | Consider AI solutions connected to your own infrastructure that don't use your data to train models | Medium |
How do you define internal policies on AI use?
An effective policy isn't a complex legal document: it's a set of clear, practical guidelines that every team member can follow without advanced technical training.
AI use policies should state clearly and unambiguously:
- which tools are approved for business use;
- which data can be uploaded to public tools;
- which activities call for extra care or controlled environments.
Documents containing customer information, financial data or strategic content should be kept out of public AI tools or go through anonymization first, meaning removing names, company references and identifying data before putting the text into a prompt.
It's also important to define clear operational responsibilities: who can use certain tools, what permissions are needed, how access should be managed. Many risky uses simply stem from a lack of practical guidance. Setting operational guidelines reduces improvised behavior and builds a more controlled approach to company data.
Staff training plays a central role, because many risks come from unaware use. According to Politecnico di Milano, only 7% of Italian SMEs have launched structured AI training programs: a figure that partly explains why Shadow AI is so widespread. To dig deeper, our guide on mistakes to avoid in your first AI project offers a useful practical checklist.
How do you separate sensitive data from public AI tools?
The distinction between data that can be shared and data that requires controlled environments is the operational foundation of any serious approach to privacy in the age of AI.
To use AI properly at work, it's essential to tell shareable information apart from information that requires controlled environments. Here are the four main operating principles.
| Principle | How to apply it | Practical example |
|---|---|---|
| Avoid uploading sensitive documents directly | Contracts, customer databases, internal price lists and financial data should not be entered into public AI platforms without prior checks | A manufacturing company doesn't upload proprietary technical specs to public ChatGPT |
| Use anonymization and summaries | Remove names, company references and identifying data from documents before using them in prompts | A professional firm replaces client names with codes before asking AI for help |
| Separate public and internal workflows | Low-risk activities (brainstorming, generic content) on public tools; processes involving strategic data in controlled environments | An e-commerce business uses public AI for copy ideas but handles CRM data only on vetted platforms |
| Evaluate private AI platforms | AI systems connected to your own digital infrastructure that don't use your data to train models | A retail chain adopts an internal AI assistant connected to its management software |

How do you vet AI software and vendors before adopting them?
Choosing AI tools requires technical and organizational checks that many SMEs underestimate. Vetting a vendor before adopting it is far easier than handling a data breach afterward.
Adopting an AI tool without vetting it is like signing a contract without reading it. Here are the essential checks to run before approving any new platform within your company infrastructure.
- 1 · IT or management leadCheck where data is storedVerify server location, the cloud infrastructure in use and how uploaded information is retained. Data of EU citizens on non-EU servers can create legal complications.
- 2 · IT or management leadReview policies and terms of useMany tools spell out in their policies how prompts, documents and user-uploaded content are handled. Ignoring this can expose your company to operational risks.
- 3 · IT or management leadCheck whether data is used for AI trainingSome platforms use the information users share to improve their models. Companies need to know exactly which data may be reused and how to turn this option off.
- 4 · IT or management leadAssess security and compliance levelsCertifications, access management, two-factor authentication, activity tracking and regulatory compliance are essential when selecting technology vendors.
A plugin that promises to summarize your emails may be able to read all your company messages in plain text, breaching every confidentiality standard. Cybersecurity isn't only about external attacks: it's also about the permissions your company grants to its own tools. If you use data analysis tools, the article on Google Data Manager and data audits with AI agents shows how to build a more controlled approach.
Three real-world scenarios: how do risks change for an e-commerce business, a professional firm and a manufacturing company?
Privacy and artificial intelligence risks show up differently depending on your industry. Understanding your own scenario helps you identify where to act first.
Illustrative examples: the cases below show how risk mechanisms work and are not real client data.
| Company type | Main risk | Data at risk | Priority action |
|---|---|---|---|
| E-commerce | Uploading customer lists to public AI tools for segmentation | Personal details, purchase history, customer emails | Use only AI tools with a signed data processing agreement |
| Professional firm (accountant, lawyer) | Entering tax or legal documents into prompts to get drafting support | Tax data, contracts, confidential client information | Always anonymize documents before using them in prompts; consider private AI solutions |
| Manufacturing company | Sharing technical specs or industrial drawings with public AI tools | Production know-how, patents, product specs | Keep workflows strictly separate: public AI only for generic content, closed environments for technical data |
Why has company data become a strategic asset, and why does protecting it pay off?
Every digital activity generates information. How well that data is collected, organized and protected determines your company's ability to read its market and make better decisions.
Between customer interactions, communication history, CRM data, campaign performance and output produced by AI systems, every company builds up an information asset worth far more than it seems. The problem is that this asset is often scattered across multiple systems and poorly governed.
| Strategic dimension | Concrete impact | Risk if left ungoverned |
|---|---|---|
| Integrating data and decisions | Reliable reports and KPIs that truly reflect performance | Decisions based on incomplete information or data that's out of sync across CRM, ERP and AI tools |
| Sales and marketing | Precise segmentation, customer profiling, more effective campaigns | Imprecise targeting, higher ad costs, lower conversion rates |
| Reducing operational risk | More predictable, controllable processes | Duplication, lost information, improper use of AI tools |
| Making the most of your information assets | Customer history and sales performance as a basis for optimizing strategy | Scattered data that produces no value or useful insight |
| Governance as a competitive advantage | Faster scaling, fewer inefficiencies, greater control | Struggling to grow without losing control of digital processes |
Companies that structure their information flows well scale faster and keep tighter control over their digital infrastructure. If you want to understand how AI can create real value, and not just costs, the analysis AI without profits: what your company is getting wrong is a useful starting point. Our guide on ChatGPT Work Data Agent for data analysis in SMEs also shows how to use AI to extract value from data in a controlled way.
How do you get started: the steps to put privacy and AI in order in your company
You don't need a long, expensive project. Six practical steps, in the right order, are enough to cut most of the risks and build a solid foundation.
Here is the path we recommend for SMEs that want to tackle the topic in a structured way, without blocking the use of AI.
- 1 · ManagementMap your existing data flowsUnderstand where company information travels: across which platforms, through which integrations, and who has access. Without this map, you can't know where to act.
- 2 · Management + ITIdentify your most sensitive dataClassify information by sensitivity: customer data, financial data, technical know-how, strategic content. Each category needs a different level of protection.
- 3 · Management + ITReview the AI tools already in useAudit the AI tools your company already uses, including unapproved ones. Check policies, server location, use of data for training and security levels.
- 4 · ManagementDefine and communicate internal policiesWrite clear, practical guidelines: which tools are approved, which data can be uploaded, who can access what. Share them with everyone on your team, including external vendors.
- 5 · HR + ManagementTrain your staff with practical guidanceSkip the complicated theory courses and use concrete examples: what people can do, what they can't, and what to do when in doubt. Training reduces Shadow AI and builds an organizational culture.
- 6 · IT + ManagementMonitor and update over timeThe AI tool landscape changes fast. Review your policies at least every six months and update training whenever new tools are introduced.
What mistakes are made most often when tackling this topic?
The most common mistakes aren't technical. They're organizational. They come from the idea that privacy is an IT department problem, not a responsibility shared across the whole company.
| Common mistake | Why it is a problem | How to avoid it |
|---|---|---|
| Delegating everything to IT without involving management | Decisions about which data to use, and how, are business decisions, not just technical ones | Management should lead the policy; IT implements it |
| Assuming the problem only affects large companies | SMEs are often more exposed because they have fewer controls and more unvetted tools | Start by mapping your data flows, even with limited resources |
| Adopting AI tools without reading their policies | The terms of use spell out how your data is handled: ignoring them is a legal and operational risk | Set aside 30 minutes to read the policies before every new tool you adopt |
| Not updating policies when your tools change | The AI landscape moves fast: a policy written a year ago may already be outdated | Review your policies and the tools in use every six months |
| Believing training is a cost, not an investment | 76% of SMEs don't invest in AI training: the result is Shadow AI and unmanaged risks | Even a two-hour session with practical guidance significantly reduces risky behavior |
| Using free AI tools for strategic data | Free software often uses inputs to train its models: the data you enter can become the vendor's property | Draw a clear line between general-purpose tools and those approved for sensitive data |
How do I know if my privacy and AI governance is working?
Data governance isn't a project with an end date. It's an ongoing process. A few concrete indicators show whether you're heading in the right direction.
| Indicator | How it is measured | Positive signal |
|---|---|---|
| Adoption of internal policies | Share of employees who know and apply the guidelines on AI use | Over 70% of employees know what they can and can't do with AI tools |
| Reduction of Shadow AI | Number of unauthorized AI tools found during audits | Steady decline compared to the initial baseline |
| AI-related security incidents | Number of reports of accidental uploads of sensitive data | Zero incidents in the three months after training |
| Vendor vetting | Share of AI tools in use with a verified policy and a signed data processing agreement | 100% of authorized tools have verified documentation |
| Policy updates | How often internal guidelines on AI use are reviewed | Documented review at least every six months |
| Staff training | Share of employees who have completed at least one training session on safe AI use | Over 90% of staff trained within 12 months of adoption |
So, is it worth investing in privacy and AI governance for your company?
It's worth it if you already use AI tools at your company (even just one), have team members who work with customer data or sensitive information, or operate in regulated sectors like healthcare, finance or legal. In these cases, not structuring your governance is a real risk: operational, legal and reputational.
A middle ground: you're considering adopting AI but haven't started yet. You can build the foundations of your governance before adopting the tools. It's far easier to do it up front than to fix things later. Start by mapping your data flows and drafting a minimal policy.
Not yet if your company doesn't use internet-connected digital tools, doesn't process third-party data and has no team members using external software. In that case the topic isn't urgent, but it's an increasingly rare scenario.
How we see it at SHM
For us, technology is the means and people are the end. That holds true here too, and especially so, when it comes to AI and privacy. We believe artificial intelligence should free companies from routine, not create new risks that someone else then has to handle in an emergency.
In our experience with Italian SMEs across different sectors, the most common problem isn't a lack of tools. It's a lack of structure. Companies use five different AI platforms, none of them vetted, with no written policy and no training. The result is a fragmented digital ecosystem that exposes data, processes and know-how without anyone noticing.
Our approach always starts with analyzing existing flows, not with adopting new tools. First we understand how information moves, then we define where and how AI can be integrated sustainably. This applies to strategic AI integration, to business data management and processing and to building custom AI assistants that work inside controlled environments.
If you want to know where to start for your company, you'll find all our practical updates on this topic in the hub dedicated to AI and process automation. For a direct conversation, we're available to support our clients with a tailored analysis of their situation.
Frequently Asked Questions
Never enter identifiable personal data, client names, contract amounts, trade secrets, proprietary source code or non-public documents. Free software often uses user inputs to train its own models: once submitted, the information could become part of the AI's knowledge base and, in theory, be served up to other users. The rule of thumb is simple: if you wouldn't publish that text on the internet, don't upload it to a public AI tool.
It refers to employees using artificial intelligence tools without management or the IT department knowing. It usually happens through personal accounts or browser extensions used to speed up everyday tasks. It's a real risk because it bypasses every company security protocol, exposing the business to potential data breaches. According to the Politecnico di Milano, it already affects 34% of Italian workers.
The AI Act introduces new transparency and security obligations for anyone who develops or uses AI systems, especially those classified as high risk. For SMEs, this means documenting processes better, monitoring system outputs and making sure the solutions they adopt comply with the new European standards. Minimal- or limited-risk applications have lighter obligations, but even then you need to ensure transparency toward users.
The geographic location of the server determines the jurisdiction the data falls under. Storing the personal data of European citizens on servers outside the EU can lead to legal complications and GDPR incompatibility. Checking that your provider is GDPR-compliant and that data is processed within a secure regulatory perimeter is an essential step before adopting any AI tool.
A ban with no alternative doesn't work: employees keep using personal tools, they just don't tell you. The most effective approach is to provide authorized, vetted AI tools, backed by clear policies and hands-on training. When team members understand why certain rules exist and have the right tools at hand, Shadow AI shrinks naturally.
Penalties vary widely depending on the severity of the breach and the size of the company. According to Federprivacy, excluding the ten highest fines of 2025, the average fine in Europe is around 298,000 euros. But the real cost also includes reputational damage, lost customer trust and the operational costs of handling the incident. In Italy, fines collected by the Garante grew by 54.5% in a single year.
The real question isn't whether to use AI, but whether your company knows where the data you feed it every day goes.
Related articles
Discover more articles exploring similar topics, selected to offer you a more complete and stimulating perspective. Each piece of content is carefully chosen to enrich your experience.