Is AI Dangerous? Honest Answers for SMB Owners
Erwin Berkouwer
AI isn't dangerous in itself, but unmanaged use is. The real risks for SMBs are privacy and the GDPR (AVG), wrong answers, too much trust without checking and AI-driven fraud. All four can be kept in check with a few clear agreements. This article gives you the facts, without hype and without fear, with concrete action points.
What are the real risks of AI for your business?
1. Privacy and the GDPR
This is the most concrete risk for Dutch companies. With some free AI tools, such as the basic version of ChatGPT, text you enter can be used to improve models unless you switch that off. If you enter customer data, quotes or personal data, that information can end up outside your control.
In practice: an employee who copies customer complaints into ChatGPT to write replies may be sharing personal data with a party outside the EU. That can breach the GDPR. The maximum fine is €20 million or 4% of worldwide annual turnover.
Tip: ask for a data processing agreement with every AI tool. European providers and business versions generally offer stronger privacy guarantees. Without a data processing agreement, you as the business owner stay responsible.
2. Wrong information (hallucinations)
AI models hallucinate: they sometimes give answers that sound logical but are factually wrong. In customer service, legal or financial situations that can have serious consequences, from bad advice to product information that misleads customers. You can read why that happens in why ChatGPT gives you bad answers.
Tip: always build in a verification step. For customer-facing or financial uses, have an employee check the output. The mistake isn't in using AI, but in trusting the output blindly.
3. Dependence without control
Anyone who leaves processes entirely to AI without human checks takes a risk if the system makes a mistake. An AI that automatically sends quotes or approves invoices without a check can make costly errors that only show up later.
4. AI fraud: deepfakes and phishing
Criminals use AI for convincing phishing emails, fake invoices and voice cloning, where a director's voice is imitated to get payments approved. This risk is growing.
Tip: set up a verification protocol for payment orders. Never carry them out based on an email or phone call alone, but always confirm through a second channel.
What is the risk of not using AI?
This is the part many articles on AI risks leave out.
- Competitive damage: companies that use AI structurally often work faster and cheaper on repetitive work, such as quotes and customer questions.
- Shadow AI: employees are already using AI anyway, just not through official and safe channels, with all the privacy risks that brings.
- Staff costs: repetitive tasks that AI does in minutes cost your team hours a week.
So the choice isn't between AI and safety. It is about use with agreements instead of use without.
How does risk management work in practice?
Example: an accountancy firm. The firm decides that employees don't enter client data into public AI tools, and chooses a secure European AI environment with a signed data processing agreement instead. That way it gets the time savings without GDPR risk.
Example: a logistics provider. A transport company uses AI for route proposals with a human-in-the-loop approach: the AI proposes a route, a planner approves it. That keeps the human responsible.
These are illustrative examples of working agreements, not measured results.
A practical risk matrix for SMBs
| Use | Risk | Recommendation |
|---|---|---|
| Writing and rewriting text | Low | Use freely, don't share customer data |
| Answering customer communication | Medium | Only use with your own data and always check |
| Financial decisions | High | Always human review, no full automation |
| Entering customer data into public tools | High | Avoid it, use secure alternatives |
| Fraud detection and security | Variable | Set up with a specialist partner |
| Processing quotes and invoices | Medium | Check the output, never send fully automatically |
How do you use AI safely?
Work with AI that uses your own data under clear agreements, and with human checks where it matters. That is how UnifyAI works with AI agents: the agent does the routine work and an employee reviews what is sensitive. You can read more about control, access and data handling on security and control.
Want to know where your business stands? An AI scan is a first step, also for a freelancer or small business.
When is this relevant for you?
This is relevant if you:
- want to use AI tools but doubt whether they're safe;
- have employees who already use AI tools on their own, through safe channels or not;
- want to know what the EU AI Act means for your business;
- need arguments to discuss it internally.
Frequently asked questions
Is ChatGPT dangerous for my business?
Not by itself, but reckless use is. Never share customers' personal data, confidential company information or passwords through public AI tools such as the free version of ChatGPT.
Can I enter customer data into ChatGPT?
In the free version that is risky from a GDPR point of view. With business versions where the provider promises not to use data for training there are more options, but sign a data processing agreement and ask your privacy adviser. More on this in using ChatGPT at work: privacy and GDPR.
What is the EU AI Act and does it apply to my business?
The EU AI Act is European legislation that sorts AI systems by risk. For most SMBs that use AI tools and don't develop them, the obligations are more limited, but the rules come into force in phases. Get advice on what applies to you.
How do I protect my business against AI fraud?
Set up a verification protocol for payment orders, train employees to recognise phishing and use two-step verification on all business accounts.
Are there safe alternatives to public AI tools?
Yes. European providers and business versions of existing tools offer stronger privacy guarantees. Always ask for the data processing agreement before you use an AI tool in your business.
What is a logical next step?
Start with a simple AI policy: which tools are allowed, which data never goes in, and where a human checks. Then run an AI scan or book an introduction to talk about how to use AI safely.
Frequently asked questions
Short, clear answers so you can decide faster.
Is ChatGPT dangerous for my business?
Not by itself, but reckless use is. Never share customers' personal data, confidential company information or passwords through public AI tools such as the free version of ChatGPT.
Can I enter customer data into ChatGPT?
In the free version that is risky from a GDPR point of view. With business versions where the provider promises not to use data for training there are more options, but always sign a data processing agreement.
What is the EU AI Act and does it apply to my business?
The EU AI Act sorts AI systems by risk. For SMBs that use AI tools and don't develop them, the obligations are more limited, but the rules come into force in phases. Get advice on what applies to you.
How do I protect my business against AI fraud?
Set up a verification protocol for payment orders, train employees to recognise phishing and use two-step verification on all business accounts.
Are there safe alternatives to public AI tools?
Yes. European providers and business versions of existing tools offer stronger privacy guarantees. Always ask for the data processing agreement before you use an AI tool in your business.
