Contents
AI governance for development teams defines which AI tools may be used for what, who carries which responsibility and how quality, security and data protection are ensured. Good governance isn't a thick rulebook but a short, readable policy with clear ownership, technical guardrails and regular review. It is the prerequisite for teams to use AI coding tools productively and without grey areas.
Why Development Teams Need Their Own AI Governance
In many companies the situation is murky: some developers use AI assistants under a company licence, others with private accounts, and some not at all because they're unsure whether they're allowed to. Coding agents run commands, install packages and open pull requests. Without clear rules, risks arise for trade secrets, privacy, security and code quality – while productivity potential goes unused.
Regulatory requirements add to this. Since 2 February 2025, Article 4 of the EU AI Act requires providers and deployers of AI systems to ensure a sufficient level of AI literacy among their staff. Further obligations of the AI Act are being phased in. What this means for training is covered in AI literacy under the EU AI Act: do companies have to train their teams? This article is not legal advice; seek qualified counsel for your specific situation.
The Seven Building Blocks of AI Governance
| Building block | Content | Typically owned by |
|---|---|---|
| Tool approval | List of approved tools and plans, process for new tools | CTO / head of IT with data protection |
| Data classification | Which repositories and data may go into which tools? | IT security, data protection officer |
| Usage rules | Prohibited inputs, handling of secrets and customer data | Tech leads |
| Quality and review | Mandatory review, tests, labelling of AI contributions | Tech leads, QA |
| Agent permissions | Sandboxing, approvals for pushes, deployments, network access | Platform / DevOps team |
| Legal and licences | Open-source licences, copyright, client contracts | Legal department or external counsel |
| Skills and training | Onboarding, regular training, documentation of training | Team leads, HR |
Sample Rules for an AI Policy
A policy for development teams should fit on a few pages and make concrete statements. These rules have proven a useful skeleton in practice – adapt them to your situation:
- Only approved tools with company accounts are used for company code.
- Credentials, keys and real personal data are never entered into prompts, chats or context files.
- Repositories classified as "strictly confidential" are only processed with tools explicitly approved for them.
- All AI-generated code is reviewed by a human before merging; whoever opens the pull request owns its content.
- Coding agents work in isolated environments; pushes to protected branches and deployments require human approval.
- New dependencies suggested by a tool are checked for origin, licence and security before being added.
- Substantial AI contributions are labelled in the pull request.
- Every team member completes an introduction to the policy and approved tools before use.
How responsibility for agent-generated code is allocated in practice is explored in agentic coding and code ownership: who is responsible for the code?
Technical Guardrails Instead of Paper Rules
Rules that exist only on paper are quickly forgotten in daily work. It's more effective to enforce as many requirements as possible technically:
- Single sign-on and central licence management so only approved accounts are used
- Admin policies in the tools, such as excluded files and directories or disabled features
- Secret scanning in pre-commit hooks and the CI pipeline
- Branch protection with mandatory reviews and green tests before merging
- Sandboxes and permission concepts for agents with minimal privileges
- Dependency and licence scanning for all new packages
- Project rules in the repository that agents read automatically, containing conventions and prohibitions
Which security risks these measures address is explained in security risks of AI-generated code and how to avoid them. For the privacy side, see AI coding tools and data privacy.
Rollout in Four Phases
- Inventory (2–4 weeks): Which tools are already in use, with which accounts, for which tasks? Honest, anonymous surveys help more than bans.
- Policy and approvals: Define the building blocks, assess and approve tools, align the policy with developers – not just with management and legal.
- Technical implementation and training: Set up guardrails, train the team, name contact persons.
- Quarterly review: AI tools change quickly. Review and adjust policy, tool list and lessons learned regularly.
The time frames are estimates and depend on company size and starting point.
Common AI Governance Mistakes
- Blanket bans: Lead to shadow AI with private accounts – and therefore more risk, not less.
- Rules that are too abstract: "Use AI responsibly" doesn't help any developer make a concrete decision.
- No developer involvement: Rules that ignore real-world practice get bypassed.
- Write once, never update: New tool capabilities such as agents with network access require new rules.
- Forgetting training: A policy without training has no effect – and the AI literacy duty remains unfulfilled.
For training your teams, we place AI developers from real-world practice as trainers – from €799 net, with content and format agreed individually.
Planning an app? Request a free fixed-price quote in 2 minutesConclusion
AI governance in development teams isn't a brake; it's the prerequisite for safe, productive use of AI coding tools. Seven building blocks – from tool approval through data classification to training – form the framework. The key is to keep rules short and concrete, enforce them technically wherever possible, involve developers and review everything regularly.
Frequently Asked Questions
What belongs in an AI policy for developers?
Approved tools, rules for confidential data and secrets, review requirements, permissions for coding agents, handling of dependencies and licences, and training requirements.
Is AI training for developers legally required?
Since February 2025, Article 4 of the EU AI Act requires providers and deployers of AI systems to ensure sufficient AI literacy among their staff. How to implement this in your case should be assessed individually.
Who is responsible for AI governance in a company?
Overall responsibility usually lies with the CTO or head of IT, supported by data protection, IT security, legal and tech leads. What matters is that ownership of each building block is clearly assigned.
How often should an AI policy be updated?
Because AI tools evolve quickly, a review roughly every quarter makes sense – and additionally whenever new tools or capabilities are introduced.