Ireland now has an AI Office, a network of sectoral regulators and an EU AI Act enforcement structure.
The next piece is the AI regulatory sandbox.
The AI Office of Ireland confirms that Ireland’s national sandbox programme is currently under development. It is intended to give organisations developing innovative AI systems a controlled environment in which to test their technology while receiving regulatory guidance.
But there is an important practical point:
As of 26 September 2026, applications are not yet open and the AI Office has not published final eligibility or application details.
For businesses developing AI, now is the time to understand whether the sandbox may eventually be relevant.
What is an AI regulatory sandbox?
An AI regulatory sandbox is not simply a technical testing environment.
Under Article 57 of the EU AI Act, sandboxes are designed to allow innovative AI systems to be developed, trained, tested and validated for a limited period within a controlled regulatory framework.
Competent authorities can provide guidance on issues such as:
- AI Act classification
- risk management
- fundamental rights
- health and safety
- testing and mitigation measures
- regulatory expectations
- compliance with other relevant EU or national law
Ireland’s AI Office describes the programme as a way for organisations to identify compliance considerations early rather than discovering them after a system is ready for market.
Who might benefit?
Ireland has not yet published its final participation criteria.
However, the AI Office says the programme is being developed for organisations including start-ups, researchers and public-sector bodies working on innovative AI solutions.
Under the wider EU AI Act framework, providers and prospective providers are the core participants in regulatory sandboxes. Applications can also involve partnerships with deployers and other relevant organisations.
This distinction matters.
A company simply using Microsoft Copilot or ChatGPT internally is unlikely to need a regulatory sandbox just because it uses AI.
A company developing an AI recruitment platform, medical AI product, financial decision system or other innovative AI solution facing genuine regulatory questions may have a much stronger case.
When will Ireland’s sandbox be operational?
The deadline has changed.
The original AI Act required Member States to establish at least one national AI regulatory sandbox by August 2026.
The 2026 Digital Omnibus amended Article 57 and moved that deadline to:
2 August 2027.
The current consolidated AI Act reflects the new date.
Ireland’s AI Office is therefore developing the programme ahead of that revised legal deadline.
This is important because older AI Act explainers and even some online regulatory resources may still display the original 2026 date.
Is the sandbox a compliance shortcut?
No.
Participating in a sandbox is designed to help organisations understand regulatory expectations and work towards compliance.
It can also generate useful evidence.
Under Article 57, participating providers can request written proof of activities completed in the sandbox and receive an exit report covering the work, results and learning outcomes.
That documentation can later support conformity assessment or market-surveillance activities.
But participation should not be treated as a regulatory stamp saying:
“this AI system is approved”.
Businesses remain responsible for meeting the legal requirements that ultimately apply to their system.
GDPR, product rules and other applicable EU and national law still need to be addressed within the sandbox.
Why SMEs should pay attention
The AI Act deliberately gives smaller companies additional support.
The sandbox framework requires access for SMEs, including start-ups, to be free of charge, except for limited exceptional costs that authorities may recover where permitted.
The framework is also intended to make application procedures understandable and accessible to smaller organisations with limited legal and administrative resources.
For an Irish AI start-up facing uncertainty about classification, risk controls or future high-risk obligations, that could be valuable.
Four things to prepare before applications open
1. Define what your AI actually does
Document the intended purpose, users, outputs and decisions the system can influence.
2. Identify the difficult regulatory questions
Do not apply simply because the product contains AI.
Ask where genuine uncertainty exists around classification, data, risk, human oversight or sector-specific regulation.
3. Prepare basic governance documentation
Have a system description, risk assessment, data-flow overview, vendor information and testing approach ready.
4. Watch the AI Office application process
The AI Office says eligibility and application details will be published once the sandbox programme has been developed.
Do not rely on unofficial application forms or outdated EU timelines.
Blanche perspective
The most valuable part of a regulatory sandbox is not permission to experiment.
It is early regulatory clarity.
For businesses developing AI, discovering a classification, data or risk-management problem shortly before launch can be expensive.
Finding the same issue while the product is still being designed is much easier to fix.
So the question is not:
“Can we get our AI approved in the sandbox?”
It is:
“Do we have a regulatory question worth testing before we scale?”
That is where the sandbox could become genuinely useful.
Sources
Primary and authoritative sources used for this Insight.
- AI Office of Ireland: Ireland’s AI Regulatory Sandbox Programme (current Irish implementation information)
- EUR-Lex: AI Act, consolidated 27 July 2026, Articles 57–58 (consolidated reference to binding EU law)
- EUR-Lex: Regulation (EU) 2026/1744, Digital Omnibus on AI (binding EU law)
- European Commission: AI Omnibus enters into force, 27 July 2026
- European Commission: 2025 sandbox implementing-act consultation (historical draft, not final binding rules)