Ireland is not a side-market. It is a small, watched, politically attentive market with a regulator that takes a position and holds it. The CRU's December 2025 decision on data centre grid connections — restricting Dublin-region connections, requiring on-site or contracted generation, permitting private wire only inside a tight envelope — was not nominally an AI decision. It was a posture decision. It told the market that the CRU is willing to constrain large customers and large infrastructure when system or consumer interests require it. AI in supply will be read against the same posture.
What the CRU will tolerate.
AI used to improve outcomes for customers — better fault detection, better tariff matching, faster vulnerable-customer identification, more accurate billing — with a clear assurance chain and named accountability. AI used to reduce regulatory friction — faster, more complete responses to information requests, better evidence on outcomes. AI used in operational forecasting and balancing in ways that respect the all-island Single Electricity Market design. The CRU has been consistent: it does not object to capability, it objects to opacity.
What the CRU will not tolerate.
AI-mediated decisions on disconnection, debt, vulnerability, prepayment or hardship that cannot be evidenced and explained. Outsourced AI risk where the supplier cannot show what its vendors and TPIs are doing with Irish customer data. Marketing or onboarding AI that quietly disadvantages protected groups. Anything that looks like a UK assurance regime stapled onto an Irish operation without recognition that ROI has its own consumer protection framework, its own data-protection regulator and its own political context. The Irish customer expectation, particularly post-2022, is for a regulator that intervenes early — and the CRU has not disappointed that expectation.
How Probari approaches the ROI surface.
Probari's regulatory agents check CRU decisions, all-island SEM rules and the supplier's own Irish-market Code of Practice as the trio that defines acceptable AI behaviour. AIIAs for ROI use cases are run against that trio explicitly, not against a GB template with a footnote. Where a supplier operates in both jurisdictions, the assurance evidence is held twice — not because the underlying AI is different, but because the regulators are.