Skip to content
Blentera

Resources, Governance

AI agent policy template for companies

An AI agent policy tells employees which agents the company uses, who is responsible for them and what they may do. Copy the outline below and fill in the parts marked in square brackets.

Updated 20 November 2026. 9 minute read.

In short

  • A useful AI agent policy has nine parts: purpose and scope, roles and owners, approved tools and models, data rules, approval rules, budgets, incident handling, register and review cycle, and training.
  • Each agent needs a named business owner, an approver with authority to stop it, and an entry in a register with a review date.
  • The policy supports EU AI Act duties that already apply, such as AI literacy (Article 4) and transparency (Article 50), and prepares for the high-risk deployer duties in Article 26 that apply to Annex III use cases from 2 December 2027.
  • Data rules should follow GDPR: a lawful basis, data minimisation, processor agreements, impact assessments for high-risk processing, and breach reporting within 72 hours where required.
  • Review the policy at least once a year and whenever the law, a model or an agent's purpose changes.

How to use this template

The outline below has nine sections. Each section gives policy text you can copy, followed by a note on how it maps to the EU AI Act (Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744) and to the GDPR (Regulation (EU) 2016/679). Replace text in square brackets with your own details.

Keep the policy short. Two to four pages is enough for most companies. Put detail, such as the list of approved tools, in appendices that the owners can update without a full policy review.

This template is a starting point, not legal advice. Have your data protection officer or legal adviser review it before adoption.

Section 1: Purpose and scope

  • Purpose: this policy sets out how [Company] selects, runs and oversees AI agents, so that they work safely, lawfully and within budget.
  • Scope: the policy covers every AI agent that acts for the company, including agents that draft, decide, send, buy or change data, whether built in-house or bought.
  • It applies to all employees, contractors and teams that set up, use or approve AI agents.
  • Out of scope: personal, non-work use of AI tools on private devices.

Mapping: the AI Act calls a company that uses an AI system under its authority a deployer (Article 3(4)). State in the policy that the company acts as a deployer, and that any change that could make it a provider under Article 25 needs approval first.

Roles, owners and approved tools

Section 2: Roles and owners

  • Policy owner: [role, for example the COO] owns this policy and the register.
  • Agent owner: each agent has one named business owner who is accountable for its purpose, results and costs.
  • Approver: each approval rule names an approver and a deputy with the competence, training and authority to reject or stop the agent's work.
  • Data protection officer or privacy lead: [name or role] reviews agents that process personal data.
  • IT or security: manages access, credentials and the connection of agents to apps and tools.

Mapping: for high-risk systems, Article 26(2) of the AI Act requires oversight by employees with the necessary competence, training and authority. Under GDPR Article 24, the controller is responsible for appropriate measures. Where a data protection officer is designated, Articles 37 to 39 apply.

Section 3: Approved tools and models

  • Only agents, platforms and AI models listed in Appendix A may be used for company work.
  • Adding a tool or model needs approval from the policy owner, IT and, where personal data is involved, the privacy lead.
  • Each approved vendor must have a data processing agreement where it processes personal data for the company.
  • Agents must be used in line with the vendor's instructions and terms. Changing an agent's purpose needs a new register entry.
  • Keep at least one alternative model for critical work, so the company can switch if a vendor changes terms.

Mapping: Article 26(1) requires deployers of high-risk systems to follow the provider's instructions for use. Article 25(1)(c) makes a deployer a provider if it turns a system to a high-risk purpose. GDPR Article 28 requires a contract with each processor, and Chapter V governs transfers of personal data outside the EU and EEA.

Data rules

Section 4: Data rules

  • Agents may only access the data needed for their task. Access is granted per app and per tool, not to whole systems.
  • Each agent that processes personal data has a recorded purpose and lawful basis.
  • Special categories of personal data, such as health data, may only be processed by an agent with written approval from the privacy lead.
  • Confidential data, trade secrets and credentials must not be entered into tools outside Appendix A.
  • A data protection impact assessment is carried out before an agent processes personal data in a way likely to result in a high risk.
  • Agent logs and memory follow the company's retention schedule.

Mapping: GDPR Article 5 sets the principles of purpose limitation, data minimisation and storage limitation. Article 6 requires a lawful basis, Article 9 restricts special categories, Article 25 requires data protection by design and by default, Article 32 requires appropriate security, and Article 35 requires an impact assessment for high-risk processing. Under the AI Act, Article 26(4) requires input data under the deployer's control to be relevant and sufficiently representative, and Article 26(9) links the deployer's assessment to the provider's information.

The European Data Protection Board's Opinion 28/2024 is useful background on how the GDPR applies when AI models are developed and used.

Approval rules and budgets

Section 5: Approval rules

  • Agents may act without approval only on internal, reversible tasks listed in their register entry.
  • An employee must approve, before it happens, any agent action that cannot easily be undone, moves money above [amount], reaches customers or the public, or shares personal data outside the company.
  • Decisions with legal or similarly significant effects on an individual, such as hiring, promotion, termination or credit, are taken by an employee. The agent may only prepare them.
  • Approvers must review the content, not just the request. Rubber-stamping is a breach of this policy.
  • Every agent can be paused at once by its owner or approver, and access can be revoked by IT.
  • Customer-facing agents identify themselves as AI.

Mapping: AI Act Article 14 requires high-risk systems to allow oversight, including a stop button or similar procedure, and Article 26(2) assigns that oversight to the deployer's employees. Article 50 sets transparency duties from 2 August 2026, including disclosure when individuals interact with an AI system. GDPR Article 22 restricts decisions based solely on automated processing that have legal or similarly significant effects.

Section 6: Budgets

  • Each agent has a monthly budget for AI usage and paid tools, set by its owner.
  • The owner is warned at 80% of the budget. At 100%, the agent pauses until the owner raises the budget or the next period starts.
  • Budget increases above [amount] need approval from [role].
  • Costs are reviewed monthly with the agent's results.

Mapping: neither the AI Act nor the GDPR sets budget rules. Budgets are an internal control. An automatic pause at the limit also works as a stop control.

Incident handling

Section 7: Incident handling

  • An incident is any agent action that causes or could cause harm, breaks this policy, exposes data or produces a seriously wrong result.
  • Any employee who sees an incident pauses the agent if they can and reports it to the agent owner and [incident contact] at once.
  • The agent owner records what happened, the actions taken and the root cause, using the activity log.
  • If personal data is involved, the privacy lead assesses whether it is a personal data breach.
  • If an agent may present a risk to health, safety or fundamental rights, the agent stays paused and the provider is informed.
  • The agent restarts only after the owner and the policy owner agree on the fix.

Mapping: under GDPR Article 33, the controller must notify the supervisory authority of a personal data breach without undue delay and, where feasible, within 72 hours, unless the breach is unlikely to result in a risk. Article 34 requires informing affected individuals when the risk is high. For high-risk AI systems, AI Act Article 26(5) requires deployers to inform the provider and the market surveillance authority and suspend use where a system may present a risk, and to report serious incidents first to the provider.

Register, review cycle and training

Section 8: Register and review cycle

  • Every agent use case is entered in the register before it goes live, with owner, purpose, data, tools, approval rules, budget, AI Act classification and review date.
  • The AI Act classification records whether the use case is prohibited, high-risk under Annex III or neither, and why.
  • No agent goes live until its entry is complete and the policy owner has signed it off.
  • Each use case is reviewed at least every [12] months, and whenever the agent, its model, its tools or its purpose change.
  • Logs are kept for at least six months, or longer where the retention schedule requires.
  • This policy is reviewed at least once a year.

Mapping: Article 26 requires deployers of high-risk systems to monitor operation and keep logs for at least six months. Article 26(7) requires employers to inform workers' representatives and affected workers before a high-risk system is used at the workplace. Article 27 requires some deployers to carry out a fundamental rights impact assessment. GDPR Article 30 requires records of processing activities, which the register can support. The high-risk duties apply to Annex III use cases from 2 December 2027.

Section 9: Training

  • All employees who use agents complete basic AI training before they start, covering what agents can and cannot do and this policy.
  • Approvers and agent owners complete role training on reviewing agent work, automation bias and stop controls.
  • Training is refreshed every [12] months and when the policy changes. Completion is recorded.

Mapping: Article 4 of the AI Act, applicable since 2 February 2025 and amended in 2026, requires providers and deployers to take measures to support the AI literacy of their staff and others who use AI systems on their behalf. Article 26(2) requires training for those who oversee high-risk systems. Where a data protection officer is designated, GDPR Article 39 includes awareness-raising and training of staff.

How Blentera helps

Blentera is a platform where companies run AI agents with employees in charge. It gives each agent a role in the org chart and lets you set approval rules per agent, project, app or tool. Budgets warn at 80% and pause the agent at the limit. Agents share one company memory, and an activity log records their actions and can be exported for incident reviews. An EU AI Act use-case register checks requirements such as human oversight and review dates before an agent goes live. It is hosted in the EU. The platform can carry out many of the rules in this policy, but the policy and meeting the law remain your company's responsibility.

Questions

Sources

  1. Regulation (EU) 2024/1689 (AI Act), consolidated text of 27 July 2026, EUR-Lex
  2. Regulation (EU) 2026/1744 (Digital Omnibus on AI), EUR-Lex
  3. Regulation (EU) 2016/679 (GDPR), EUR-Lex
  4. Navigating the AI Act: frequently asked questions, European Commission
  5. Article 26: Obligations of deployers of high-risk AI systems, AI Act Service Desk
  6. Opinion 28/2024 on data protection aspects of processing personal data in the context of AI models, EDPB
  7. Council gives final green light to simplify and streamline AI rules, Council of the EU, 29 June 2026

This guide is general information, not legal advice. Check how the rules apply to your company with your own advisers.

Get started

Put your first agent to work in minutes.

Three steps, and your first agent is working for your company.

  1. Step 1

    Create an account

    Sign up with your work email and invite your team. No sales call needed.

  2. Step 2

    Add your first agent

    Create an agent, give it a role, and connect your own AI key and the apps it may use.

  3. Step 3

    Give it a task

    Ask in Blentera or your chat app, and follow the work until it is done.