Company AI policy and the AI Act: clear rules for using AI at work
Yellow Tech writes the company’s AI policy together with it: which tools are used, which data is never entered, who approves a new use. It trains staff as article 4 of the AI Act requires and builds AI agents with human oversight.Yellow Tech writes the company’s AI policy together with it: which tools are used, which data is never entered, who approves a new use. It trains staff as article 4 of the AI Act requires and builds AI agents with human oversight.
What is a company AI policy?
A company AI policy is the document with the rules for using artificial intelligence at work: allowed tools, data not to enter, who approves a new use.
It starts from what people already use, often with personal accounts, and from the risk of each use under the AI Act.
An AI policy and the map of uses, with the risk of each one
Example
The policy
Company AI policy, version 1.2
Officine Rivalta
Approved tools
ChatGPT Enterprise and Microsoft Copilot, with the company account.
Personal accounts are not allowed.
Data not to enter
Customer data.
Employees’ health data.
Confidential contracts and prices.
Who approves a new use
The IT manager, through a request form.
Review
Every six months, or when the regulation changes.
Read by the company’s legal counsel
Map of uses and risk level
Writing emails and quotes
ChatGPT
MinimalSummarising meetings
Copilot
MinimalAssistant on the website
Custom agent
LimitedIt must be disclosed that it is an AI.
Reading delivery notes and invoices
Custom agent
MinimalCV screening
To be assessed
HighUse suspended until the assessment is done.
AI Act training, art. 4
48people trained
- Attendance register
- Verification test
How a company AI policy is made
Interviews with managers
We gather how AI is used today, with which tools and data.We gather how AI is used today, with which tools and data.
The map of uses
Each use with its tool and its risk.Each use with its tool and its risk.
The written policy
Allowed tools, data not to enter, who approves a new use.Allowed tools, data not to enter, who approves a new use.
The lawyer’s review
The company’s lawyer or DPO reviews the text.The company’s lawyer or DPO reviews the text.
Staff training
The article 4 course, with a record and a test.The article 4 course, with a record and a test.
What the AI Act asks today of companies that use AI
Training staff
Since February 2, 2025 whoever uses AI systems must take measures so that staff have the skills to use them: this is AI literacy.Since February 2, 2025 whoever uses AI systems must take measures so that staff have the skills to use them: this is AI literacy.
Prohibited uses
Since February 2, 2025 some uses are prohibited, for example recognizing employees’ emotions.Since February 2, 2025 some uses are prohibited, for example recognizing employees’ emotions.
Saying it is an AI
From August 2, 2026, whoever uses a chatbot with customers or publishes generated content must disclose it (article 50).From August 2, 2026, whoever uses a chatbot with customers or publishes generated content must disclose it (article 50).
High-risk systems
For high-risk uses, such as recruitment, the rules apply from December 2, 2027; for AI inside regulated products from August 2, 2028.For high-risk uses, such as recruitment, the rules apply from December 2, 2027; for AI inside regulated products from August 2, 2028.
Where the data runs: cloud or on-premise
Yellow Tech builds agents in four ways, chosen from the data the agent handles.Yellow Tech builds agents in four ways, chosen from the data the agent handles.
- Where the data is
- On the vendor’s servers, such as ChatGPT Enterprise or Claude Enterprise
- Who runs it
- The vendor
- When it is chosen
- To start right away
- Where the data is
- On servers in the European Union
- Who runs it
- Yellow Tech
- When it is chosen
- To avoid running the technology
- Where the data is
- In the company’s cloud, such as AWS, Azure or Google Cloud, inside its perimeter
- Who runs it
- The company, with Yellow Tech’s oversight
- When it is chosen
- For companies with a cloud and rules on data
- Where the data is
- On the company’s own servers, even cut off from the internet, with open models such as Llama, Mistral or Qwen
- Who runs it
- The company, with Yellow Tech’s oversight
- When it is chosen
- For healthcare, public administration and anyone who cannot use the cloud
| Where the data is | Who runs it | When it is chosen | |
|---|---|---|---|
| The AI vendor’s cloud | On the vendor’s servers, such as ChatGPT Enterprise or Claude Enterprise | The vendor | To start right away |
| SuperAgent, Yellow Tech’s platform | On servers in the European Union | Yellow Tech | To avoid running the technology |
| The company’s cloud | In the company’s cloud, such as AWS, Azure or Google Cloud, inside its perimeter | The company, with Yellow Tech’s oversight | For companies with a cloud and rules on data |
| On-premise | On the company’s own servers, even cut off from the internet, with open models such as Llama, Mistral or Qwen | The company, with Yellow Tech’s oversight | For healthcare, public administration and anyone who cannot use the cloud |
The tools the policy covers
ChatGPT
Plan, account and data not to enter.Plan, account and data not to enter.
Microsoft Copilot and Microsoft 365
What it can read in company documents.What it can read in company documents.
Claude
Account and documents to upload.Account and documents to upload.
Gemini
The same rules.The same rules.
Perplexity
What can be written in the searches.What can be written in the searches.
Mistral
The same rules.The same rules.
Custom AI agents
Human oversight, an action log, data where the company decides.Human oversight, an action log, data where the company decides.
AI inside department software
The AI features of CRMs, ERPs and HR tools, often already on.The AI features of CRMs, ERPs and HR tools, often already on.
Personal accounts
The policy says whether and when they are allowed.The policy says whether and when they are allowed.

The AI policy with the governance guidelines and the choice of tools before rolling out AI, with a measure of the AI maturity of its 60 people.
Read the story
Malta-based B2B off-price trading company, reselling apparel, footwear, accessories and home décor from over 250 international brands.
- Industry
- B2B off-price trading
- People
- 51-200
Frequently asked questions about AI policy and the AI Act
The AI Act does not ask for a document with that name, but it asks companies to train staff and to respect prohibitions and transparency duties: a written policy is the simplest way to show what the company does. The GDPR still applies to all personal data.The AI Act does not ask for a document with that name, but it asks companies to train staff and to respect prohibitions and transparency duties: a written policy is the simplest way to show what the company does. The GDPR still applies to all personal data.
Allowed tools, data not to enter, who approves a new use, the map of uses with the risk and how often it is reviewed.Allowed tools, data not to enter, who approves a new use, the map of uses with the risk and how often it is reviewed.
Read the guide and download the templateIt asks companies to take measures so that staff have the skills to use AI (AI literacy). After the 2026 Digital Omnibus what counts is being able to show what was done, for example with an attendance record and a test.It asks companies to take measures so that staff have the skills to use AI (AI literacy). After the 2026 Digital Omnibus what counts is being able to show what was done, for example with an attendance record and a test.
Discover AI Academy trainingPenalties reach up to 7% of worldwide annual turnover for prohibited uses and up to 3% for the other obligations, or a fixed amount if higher; for SMEs the lower of the two applies. In Italy the authorities named by law 132 of 2025 supervise.Penalties reach up to 7% of worldwide annual turnover for prohibited uses and up to 3% for the other obligations, or a fixed amount if higher; for SMEs the lower of the two applies. In Italy the authorities named by law 132 of 2025 supervise.
The company’s lawyer or DPO: Yellow Tech writes the policy and hands it over for review.The company’s lawyer or DPO: Yellow Tech writes the policy and hands it over for review.
With a person’s oversight where needed, a log of every action and data where the company decides. For personal data Yellow Tech is the data processor and signs a DPA, and the company remains the controller.With a person’s oversight where needed, a log of every action and data where the company decides. For personal data Yellow Tech is the data processor and signs a DPA, and the company remains the controller.
Discover AI FactoryYellow Tech is an Italian company that writes the AI policy with companies, trains staff as article 4 of the AI Act requires and builds AI agents with oversight and a log.Yellow Tech is an Italian company that writes the AI policy with companies, trains staff as article 4 of the AI Act requires and builds AI agents with oversight and a log.
If it uses AI systems it must train the staff who use them, avoid prohibited uses and, from August 2, 2026, disclose when a customer is talking to an AI or reading generated content. For high-risk uses, such as recruitment, the rules arrive from December 2, 2027.If it uses AI systems it must train the staff who use them, avoid prohibited uses and, from August 2, 2026, disclose when a customer is talking to an AI or reading generated content. For high-risk uses, such as recruitment, the rules arrive from December 2, 2027.
Neither makes a use compliant on its own: with on-premise personal data stays on the company’s servers, in the cloud it depends on the vendor and where its servers are. The AI Act looks at how AI is used, not at where the model runs.Neither makes a use compliant on its own: with on-premise personal data stays on the company’s servers, in the cloud it depends on the vendor and where its servers are. The AI Act looks at how AI is used, not at where the model runs.
It depends on how many departments are interviewed, how many uses of AI are mapped, the article 4 training and any custom agents to include. The quote comes after a first analysis of the process.It depends on how many departments are interviewed, how many uses of AI are mapped, the article 4 training and any custom agents to include. The quote comes after a first analysis of the process.
Is your team already using AI without written rules?
Yellow Tech writes your AI policy with you and trains staff as the AI Act requires.





















