Short answer
For most engineering teams, AI coding tools are not high-risk AI systems under the EU AI Act, and the team is a deployer rather than a provider. The main obligation that applies today is AI literacy under Article 4, which requires sufficient training for staff who use the tools. The picture changes when your team builds AI features into products, which can make your organization a provider with its own duties.
On this page
- The short version for engineering leaders
- Roles: who is who when your team uses a coding agent
- AI literacy: the duty that already applies
- When engineering becomes a provider
- Timeline at a glance
- Governance that satisfies the AI Act and good engineering
- Where an orchestration platform helps
- FAQ
- Tools from the jocapps ecosystem
The short version for engineering leaders
The EU AI Act (Regulation (EU) 2024/1689) regulates AI systems by risk and by role. AI coding tools such as Claude Code, OpenAI Codex, Cursor or GitHub Copilot help people write software. In almost all setups they do not fall into the prohibited or high-risk categories, because writing code is not one of the use cases listed in Annex III, and the tools are not safety components of regulated products.
That does not mean "nothing to do". Three things matter:
- AI literacy (Article 4) has applied since 2 February 2025 to providers and deployers of all AI systems, including coding assistants.
- Your role can shift from deployer to provider when your team ships AI features to customers or users.
- Other laws apply to the code and data regardless of the AI Act: GDPR, copyright and licence obligations, and product security rules such as the Cyber Resilience Act.
This article is an orientation for engineering teams, not legal advice. Involve your legal and compliance functions for your specific situation.
Roles: who is who when your team uses a coding agent
The AI Act assigns obligations by role. For a typical setup:
| Actor | Role under the AI Act | Typical obligations |
|---|---|---|
| Company behind the model (for example Anthropic, OpenAI, Google) | Provider of a general-purpose AI model | Technical documentation, copyright policy, training data summary; additional duties for models with systemic risk |
| Vendor of the coding tool | Provider of the AI system | Depends on risk class; for coding tools mostly transparency and general duties |
| Your organization using the tool internally | Deployer | AI literacy; follow instructions for use; extra duties only if the use is high-risk |
| Your organization shipping an AI feature in its product | Provider of that AI system | Obligations according to the risk class of that system |
The key distinction: using an AI tool to write code makes you a deployer of that tool. Shipping an AI system to others, under your name, makes you a provider of that system. A team can be both at the same time.
AI literacy: the duty that already applies
Article 4 requires providers and deployers to take measures to ensure, to their best extent, a sufficient level of AI literacy among staff and other people dealing with AI systems on their behalf, taking into account their technical knowledge, experience, education and the context of use.
The Act does not prescribe a certificate or a course format. For engineering teams, a defensible programme covers:
- How the tools work and fail. Why agents produce plausible but wrong code, invent APIs or packages, and how context files such as
CLAUDE.mdandAGENTS.mdshape output. - Data handling. Which code, data and secrets may be sent to which tool, and what the vendor contracts say about retention and training.
- Review duties. That AI-generated code is reviewed like any other code, with extra attention to security, licences and tests.
- Permissions and autonomy. What agents may execute, which repositories and environments they can touch, and who approves.
- Escalation. Where to report incidents, such as leaked secrets or unexpected agent actions.
Document what you did: who was trained, when, on what, and how you keep it current as tools change. Hands-on formats stick better than slide decks; neooutput runs build-as-you-learn workshops on agentic coding where every session ends with working results on the team's own code, which also gives you a record of the training. See our guide to agentic coding workshops for enterprises.
When engineering becomes a provider
The more relevant AI Act question for many software teams is not the coding assistant but what they build with it. If your product includes an AI feature, you may be the provider of an AI system. Then ask:
- Is the use case listed as high-risk? Annex III covers areas such as employment and worker management, access to essential private and public services, education, critical infrastructure, law enforcement and migration. An AI feature that ranks job applicants or evaluates creditworthiness lands here; a code completion tool does not.
- Does a transparency obligation apply? Article 50 requires, among other things, that people are informed when they interact with an AI system such as a chatbot, unless obvious, and that synthetic content is marked in a machine-readable way by providers of generating systems. These obligations apply from 2 August 2026.
- Did you substantially modify someone else's high-risk system, or put your name on it? Under Article 25 you can take on provider obligations that way.
Where you are a provider of a high-risk system, the obligations include a quality management system that covers design and development practices. At that point, how your team uses coding agents (review rules, testing, traceability of changes) becomes part of what you must be able to document.
Timeline at a glance
| Date | What applies |
|---|---|
| 1 August 2024 | AI Act enters into force |
| 2 February 2025 | Prohibited practices and AI literacy (Article 4) |
| 2 August 2025 | Obligations for general-purpose AI model providers; governance and penalties framework |
| 2 August 2026 | Most remaining provisions, including Annex III high-risk systems and Article 50 transparency |
| 2 August 2027 | High-risk AI in products covered by EU harmonisation legislation (Annex I) |
In late 2025 the European Commission proposed changes that would postpone parts of the high-risk timeline. Check the current legal status before planning around specific high-risk dates.
Governance that satisfies the AI Act and good engineering
The practices that make AI coding safe are largely the same ones that make your AI Act position defensible:
- An approved tools list with the contractual basis for each (data processing terms, retention, training opt-outs, data location). Our article on writing an AI coding tool policy gives a template structure.
- Data classes that define which repositories and data may be used with which tools. For sensitive material, consider self-hosted coding agents or anonymization. Textwash Pro removes personal data from text locally before it reaches external models.
- Traceability. Mark agent-authored commits or pull requests, keep agent session logs where tools allow, and retain review records.
- Permission boundaries. Agents run with least privilege; production credentials stay outside agent environments.
- Training records for AI literacy, refreshed when tools or policies change.
Our guide to security and data privacy of AI coding tools covers the GDPR side in detail.
Where an orchestration platform helps
Once AI moves beyond individual developer tools into business processes, governance becomes an operational problem: which agents exist, what data they touch, who approved them, what they did. kalAIDoCreate is an enterprise and public-sector AI orchestration platform with an expert team: governed AI agents, audit trails, SSO, integrations with systems such as SAP, CRM, SharePoint and Slack, and EU data residency. We build AI agents that take responsibility: each logs every action and escalates when unsure, while humans keep accountability for decisions with legal effect, which produces the documentation the AI Act expects. If your engineering team is starting to ship AI features, talk to the kalAIDoCreate team early, before role and risk classification questions arrive from customers or auditors.
FAQ
Is GitHub Copilot or Claude Code a high-risk AI system under the EU AI Act?
In typical use for writing software, no. Software development is not an Annex III high-risk use case, and coding assistants are not safety components of regulated products. Your organization is usually a deployer of these tools.
What does AI literacy under Article 4 require for developers?
Measures to ensure staff using AI tools have sufficient understanding of how they work, their risks and how to use them responsibly in context. The Act does not prescribe a format; documented, role-specific training is the practical answer.
Does the AI Act regulate AI-generated code?
Not as such. Code is not an AI system just because an AI wrote it. Obligations arise if that code is part of an AI system you provide, and other laws on security, data protection and licensing apply anyway.
Are there fines for missing AI literacy measures?
Article 4 does not carry its own specific fine in the Act, but national authorities can consider it, and missing training weakens your position in any incident. Treat it as a real obligation.