The Thomson Reuters Foundation recently released the first findings from its AI Company Data Initiative (AICDI), an analysis of AI governance disclosures from 1,000 companies across 13 industries. Developed in partnership with UNESCO, AICDI is currently the world’s largest dataset tracking how corporations describe, manage, and oversee their use of artificial intelligence.
The results point to a widening gap between AI deployment and governance. While companies are rolling out AI systems at speed, their ability to manage the associated risks—particularly environmental ones—has lagged significantly behind.
According to the analysis, 97% of surveyed companies did not assess the environmental impact of their AI systems, including energy consumption or carbon emissions. Although many firms describe their AI as “ethical,” “trustworthy,” or “secure,” very few link AI deployment to rising electricity demand, emissions, or broader climate commitments.
This disconnect, experts say, reflects a narrow view of AI governance that prioritises policy frameworks over operational realities.
Exergio, a company specialising in energy efficiency tools for real estate, argues that organisations should evaluate AI not only for ethical and regulatory risks, but also for its impact on energy performance.
“Many people assume AI will always waste energy, so they never stop to ask about its environmental impact,” said Donatas Karčiauskas, Chief Executive Officer of Exergio. “But that’s not true. There are AI tools that do the opposite—they cut consumption. Advanced building management systems, for example, use AI to lower heating and cooling demand instead of increasing it.”
Karčiauskas said the governance gap is most visible in asset-heavy sectors such as commercial buildings, where AI systems directly control physical infrastructure. In these environments, poorly governed AI can lead not only to compliance risks, but also to substantial, unmeasured energy waste.
The report also highlights weaknesses in internal AI governance. While 76% of companies with an AI strategy say oversight exists at the management level, only 41% make their AI policies accessible to employees or require staff to formally acknowledge them.
“This points to a deeper problem in how AI impact is evaluated, even at a regulatory level,” Karčiauskas said. “If you don’t observe what AI is doing in real time, you’re guessing whether it helps or harms your objectives.”
In building operations, he added, effective AI governance requires detailed operational data—such as when systems activate, how much power they consume, and what changes after AI is deployed. “Without that level of measurement, AI governance is reduced to paperwork.”
When integrated properly into building management systems, AI can smooth demand peaks, prevent unnecessary overlap between heating and cooling equipment, and reduce run times for boilers and chillers. These efficiencies, however, remain largely invisible if energy performance is not tracked alongside AI use.
Regionally, the AICDI data shows that companies in Europe, the Middle East, and Africa lead in publishing AI strategies, with 53% reporting one—largely driven by regulatory pressure from the EU AI Act. Even so, environmental impacts such as energy use and emissions are rarely included in AI governance disclosures, including within Europe.
“Europe is ahead on regulation, but even here the energy footprint of AI is mostly missing from the conversation,” Karčiauskas said. “As the EU AI Act evolves, operational transparency—how much power AI uses and whether it reduces consumption—needs to be part of governance. Otherwise, it becomes too easy to claim ‘responsible AI’ on paper while ignoring real-world energy outcomes.”
The AICDI findings suggest that most companies are still in the early stages of understanding where AI is deployed within their operations and how it affects environmental performance. Without more systematic measurement, researchers warn, AI’s growing role in business may continue to expand faster than the frameworks designed to manage its real-world impact.
