AI awareness training for employees: AI security, AI compliance and shadow AI

In brief: AI awareness training teaches employees to recognize attacks that use AI, such as AI-written phishing and voice cloning (AI security awareness), and to use AI tools at work without leaking data or breaking rules (AI compliance). Since 2 February 2025, article 4 of the EU AI Act requires organizations that use AI systems to take measures for the AI literacy of their staff. A good program covers both sides, rests on a clear AI policy and keeps records of who was trained.

AI brings about as many opportunities as risks. Attackers use it to write better phishing and to fake voices, while employees paste customer data into public chatbots to save ten minutes.

AI awareness, AI security awareness and AI compliance training compared

Many organizations start with security, because that threat is direct. Compliance is just as urgent: many AI incidents start not with a hacker but with a well-meaning employee.

How AI changes the threat landscape

AI-written phishing and social engineering

Phishing used to give itself away with clumsy language and generic greetings. Generative AI removes both: a model can read someone’s LinkedIn profile and write an email in the tone of a colleague, in flawless Dutch or German, at any scale. Microsoft measured the effect in its Digital Defense Report 2025: AI-automated phishing reached a click-through rate of 54 percent, 4.5 times higher than traditional phishing. Looking for spelling mistakes no longer works. Judging the request itself does: is it unexpected, urgent, about money or login details? Our guide what is phishing covers the basics.

Malware and vulnerability research with chatbots

Writing malware or researching vulnerabilities used to require skilled hackers. In October 2024 OpenAI confirmed that it had disrupted more than 20 malicious operations that abused ChatGPT, among other things to debug malware and prepare spear phishing. It gave attackers no new capabilities, but made them faster. The result for employees: more attacks in their inbox, also from less skilled criminals.

Deepfakes and voice cloning

A deepfake is a realistic fake video or audio recording of a real person. In 2024 an Arup employee in Hong Kong transferred about US$25.6 million after a video call in which the CFO and other colleagues were all deepfakes, CNN reported. That same year criminals tried to impersonate WPP chief executive Mark Read with a voice clone in a Teams meeting, which failed because staff stayed alert. In February 2025 a cloned voice of Italian defence minister Guido Crosetto persuaded businessman Massimo Moratti to transfer almost one million euros. The money was traced to a Dutch bank account and frozen. More cases are collected in our deepfake examples.

The same trick is used on private people: a call in which you hear your daughter’s panicked voice asking for money. Would you hesitate? Most people would not. Doubting a familiar voice is exactly the reflex training has to build.

Shadow AI: the risk that starts inside

Shadow AI is the use of AI tools that the organization has not approved or cannot see: a free chatbot on a personal account, a browser extension that summarizes emails, an AI note taker in meetings. The risk is what goes in. Free versions may store what you type and use it for training, and a pasted customer file cannot be taken back.

In IBM’s Cost of a Data Breach Report 2025, one in five organizations studied reported a breach involving shadow AI. For organizations with a lot of shadow AI, a breach cost on average US$670,000 more than for organizations with little or none. Only 37 percent of organizations had policies to manage AI or detect shadow AI.

The Dutch Data Protection Authority received dozens of reports in 2024 and 2025 of sensitive information shared through chatbots. At the municipality of Eindhoven, an internal sample showed that employees had entered confidential documents into public chatbots over thirty days, including youth care files, Accountant.nl reported in December 2025. The municipality has since blocked public AI websites.

Blocking alone rarely solves it, because AI saves people time. The lasting fix is a clear policy, an approved alternative and training that explains why the rules exist.

Why classic security awareness training is not enough

Many programs still teach people to spot typos in emails that no longer contain typos, and say nothing about what may be typed into a chatbot. AI security awareness training zooms in on AI threats, so that employees recognize and report them. That matters because most of us overestimate how well we spot threats, until the IT manager runs a phishing simulation and we enter our credentials.

AI compliance training and the EU AI Act

The main legal driver in Europe is the EU AI Act, Regulation (EU) 2024/1689. For training, article 4 is the one that counts. Since 2 February 2025 it requires providers and deployers of AI systems to take measures to support the AI literacy of staff who work with AI on their behalf. An organization whose employees use tools such as ChatGPT or Copilot for work generally counts as a deployer. The Digital Omnibus on AI, in force since July 2026, kept the duty but dropped the requirement to reach a “sufficient” level, according to the European Commission. National authorities enforce from 2 August 2026. No certificate is needed: internal training records are enough. What AI literacy means in practice is explained on our page about AI literacy training.

AI Act fines go up to €35 million or 7 percent of worldwide turnover, but that ceiling is for banned practices. Article 99 does not list article 4 among the fined obligations. The real risk: after an incident, an organization that cannot show it trained its people has a lot to explain.

AI compliance training makes staff and management familiar with topics such as:

• Risk categories: recognizing whether an AI system is banned, high-risk or neither
• Transparency: telling people when they are dealing with AI or with AI-generated content
• Privacy: what the GDPR means for personal data in prompts, uploads and AI output
• Human oversight: checking AI output before it is used in decisions, and keeping an audit trail

An AI acceptable use policy: what it should contain

Training without a policy leaves employees guessing, and a policy without training ends up in a drawer. An AI acceptable use policy sets the rules in plain language. A workable one fits on two pages and covers at least:

• Approved tools: which AI tools may be used, with which account, and how to request a new one
• Data rules: which data may never go in (personal data, customer files, source code, contracts, passwords) and what may go into approved tools
• Checking output: AI output is a draft, and a person stays responsible for facts, figures and tone
• Prohibited uses: for example decisions about people without human review, or AI note takers in confidential meetings
• Reporting: where to report a mistake, such as data pasted into the wrong tool, without fear of blame
• Ownership and review: who owns the policy and how often it is updated

Training makes the policy stick by turning rules into situations. “Do not enter personal data” is abstract. A colleague pasting a customer complaint, address included, into a free chatbot is not. Repeat such scenarios and the policy becomes habit. That is the core of a security awareness culture.

What an AI awareness training curriculum covers

How to roll out AI training for employees

1. Map AI use. Find out which tools are actually used, by whom and for what. A short anonymous survey often reveals more than network logs.
2. Write or update the AI policy and offer an approved alternative for the most common use cases.
3. Set a baseline. Run a phishing simulation and a short knowledge check before the training starts.
4. Train continuously in short sessions. A yearly e-learning fades quickly, while short and regular microlearning keeps knowledge fresh.
5. Tailor by role: deepfake payment requests for finance, fake candidates for HR, secrets in AI tools for developers, governance for management.
6. Document everything: who was trained, when and on what, plus the version of the policy. That is your evidence under the AI Act and in NIS2 and ISO 27001 audits.

Make sure the program you choose includes an AI module. Guardey has one: in short weekly challenges of about three minutes, employees learn about the latest AI threats, such as deepfakes. Phishing simulations are included, teams compete on leaderboards and auditors accept the reporting for NIS2, ISO 27001 and more than 40 other frameworks.

How to measure whether AI awareness training works

Completion rates show clicks, not behavior. Measure behavior instead:

• Phishing simulation results: the click rate and, more important, the report rate, tracked over time
• Shadow AI signals: traffic to unapproved AI tools and data loss prevention alerts, before and after training
• Self-reported mistakes: more reports is often a good sign, because people dare to speak up
• Participation over time: steady weekly participation says more than one peak after a mandatory session

Review the figures every quarter, next to the new AI tools in use and the new attack techniques. A program that is not updated several times a year falls behind the attackers it is meant to stop.

Frequently asked questions

What is AI awareness training?

Training that teaches employees how criminals use AI in attacks such as deepfakes, and how to use AI tools at work safely and within the rules. Usually as short, recurring modules.

What is the difference between AI security awareness training and AI compliance training?

AI security awareness training is about defending against attacks that use AI, such as phishing, deepfakes and voice cloning. AI compliance training is about your own use of AI: approved tools, data protection, the EU AI Act and the company AI policy. Most organizations need both.

Is AI training for employees mandatory under the EU AI Act?

Since 2 February 2025, article 4 requires organizations that use AI systems to support the AI literacy of their staff. The law prescribes no specific course or certificate, but training with records is the usual way to meet it.

What is shadow AI?

Shadow AI is the use of AI tools that the organization has not approved, often free chatbots on personal accounts. The main risk is that confidential or personal data ends up in tools the organization does not control.

Is AI-powered security awareness training the same as AI awareness training?

No. In AI-powered training the platform itself uses AI, for example to personalize content. AI awareness training is about AI as a topic. A platform can do both.