Risks of AI integration and policy recommendations

This report by Ember examines the risks associated with integrating artificial intelligence (AI) into the power sector, specifically within the ASEAN region. It details technical, operational, and governance challenges—including data limitations, cybersecurity vulnerabilities, and regulatory gaps—while providing policy recommendations to ensure a safe and efficient energy transition.

• AI integration in the power sector is hindered by significant data limitations, including fragmented energy systems, non-standardised data, and a lack of high-quality datasets. These gaps increase the risk of unintentional AI failures, such as bias, extrapolation errors when models encounter conditions outside their training, and model misalignment where AI actions diverge from operator goals.
• The growth of AI is driving an increase in data centre demand, which stresses existing power grids. By 2030, data centres could represent between 2% and 30% of national electricity demand across ASEAN countries, excluding Viet Nam. Because renewables currently only partially meet the need for stable and continuous loads, many of these facilities rely on natural gas, which increases emissions.
• Cybersecurity readiness is uneven across the ASEAN region, posing risks to regional data security and cross-border energy collaboration. While Singapore and Malaysia have high capacity and the Philippines issued a National Cybersecurity Plan in 2024, countries such as Cambodia, Myanmar, and Lao PDR remain below the world average. Cyber threats are intensifying; in 2024, a typical gas and electricity utility faced over 1,500 attacks per week, which is triple the volume from four years prior.
• Integrating probabilistic AI models into traditionally deterministic power systems creates regulatory and liability uncertainty. Current grid codes and regulations were not designed for AI decision-making, making it difficult to attribute responsibility for system failures among utilities, operators, developers, and vendors. To mitigate this, the report suggests 'human-in-the-loop' systems and the use of explainable AI to improve auditability and trust.
• To safely scale AI, the report recommends several policy actions: aligning regulations to incentivise accurate variable renewable energy (VRE) forecasting, establishing secure collaborative data ecosystems, and implementing 'AI sandboxes' for controlled testing. It also emphasizes the need for targeted workforce upskilling to reduce organisational resistance and the adoption of Machine Learning Operations (MLOps) to detect adversarial data manipulation.

APA

Ember (2026). Risks of AI integration and policy recommendations. https://ember-energy.org/chapter/risks-of-ai-integration-and-policy-recommendations/

Chicago

Ember. Risks of AI integration and policy recommendations. 2026. https://ember-energy.org/chapter/risks-of-ai-integration-and-policy-recommendations/.

Wikipedia

{{cite report |author=Ember |title=Risks of AI integration and policy recommendations |date=3 March 2026 |url=https://ember-energy.org/chapter/risks-of-ai-integration-and-policy-recommendations/ |access-date=17 August 2026 |via=Climate Insights Directory}}

BibTeX

@techreport{ember2026risks, author = {{Ember}}, title = {{Risks of AI integration and policy recommendations}}, institution = {Ember}, year = {2026}, month = mar, url = {https://ember-energy.org/chapter/risks-of-ai-integration-and-policy-recommendations/}, urldate = {2026-08-17}, note = {Indexed by Climate Insights Directory} }