Powering Trust: Industry Unites on AI Guardrails for the Electric Grid

PALO ALTO, CA – August 06, 2026

The energy sector is facing a profound paradox. The very artificial intelligence that promises to revolutionize grid management, optimize renewable energy integration, and enhance reliability is also driving an unprecedented surge in electricity demand. As data centers powering AI models are projected to consume nearly a fifth of U.S. electricity by 2030, the question is no longer if AI will transform the power grid, but how the industry can ensure that transformation is safe, reliable, and trustworthy.

Answering that question is the goal of a new, landmark initiative. The Electric Power Research Institute (EPRI) today announced the launch of SAFERai.power, a collaborative effort designed to create the essential guardrails for deploying AI in one of society's most critical infrastructure sectors. Building on the work of its Open Power AI (OPAI) consortium, EPRI is uniting utilities, technology behemoths, and regulatory bodies to build a common framework for managing the operational risks of AI.

“AI is ready to transform how the power system operates—but only if we make it trustworthy in the real world,” said EPRI President and CEO Arshad Mansoor. “With SAFERai.power, we are building the bridge between innovation and trust, applying guiding principles to the tools needed for operations. Working together, we can ensure these powerful capabilities are deployed in a way that is safe and reliable for everyone.”

The initiative’s name is an acronym for its core tenets: ensuring AI systems uphold Safety, Accountability, Fairness, Explainability, and Reliability. It moves beyond abstract principles to create tangible tools that can be used in the high-stakes environment of utility operations.

A United Front Against Fragmentation

Perhaps the most significant aspect of SAFERai.power is the sheer breadth of its coalition. More than 25 founding members have joined the effort, representing a cross-section of the entire energy ecosystem. The roster includes major utilities like Duke Energy, Exelon, and Pacific Gas & Electric; technology giants Microsoft and NVIDIA; reliability organizations like the North American Electric Reliability Corporation (NERC); and regulatory bodies represented by the National Association of Regulatory Utility Commissioners (NARUC).

This kind of collaboration is unprecedented but essential. In the absence of a shared standard, individual utilities and technology providers have been forced to invent their own risk assessment methods, leading to a fragmented, inefficient, and potentially inconsistent approach to safety. SAFERai.power aims to replace this patchwork with a unified, open-source foundation.

The participation of tech leaders like Microsoft underscores the shared responsibility in deploying AI within critical infrastructure. “Trust is the foundation for AI adoption in the power industry. Utilities operate some of society's most critical infrastructure, where reliability, security, and safety are non-negotiable,” stated Darryl Willis, corporate vice president of Energy and Resources at Microsoft Frontier Company. His sentiment highlights a crucial shift: the recognition that for AI to succeed, its creators must work hand-in-glove with its deployers to build solutions that are transparent and grounded in deep industry expertise.

This collaborative spirit extends to the regulatory sphere. Rather than waiting for rules to be imposed, the industry is proactively building the framework to inform future policy. “State regulators support innovations that can allow utilities to strengthen the reliability, resilience, affordability, and security of the electric system,” said NARUC President Ann Rendahl. “Initiatives such as SAFERai.power can help provide the collaborative framework and best practices needed to advance adoption of these technologies.”

From Principles to Practice: Building the Toolkit

While the collaboration is the engine, the initiative's true impact will be measured by its tangible outputs. SAFERai.power is not a whitepaper or a set of recommendations; it's a development project with a clear, 18-month roadmap to deliver practical tools for engineers and operators. The effort is focused on three key deliverables.

First is a Common Operational Risk Framework tailored specifically for the power sector. This framework will go beyond generic AI ethics to define concrete risk scenarios, failure modes, and guardrails relevant to grid operations. It will help a utility classify the risk of an AI application based on its use case, its level of autonomy, and the potential consequences of failure.

Second, the initiative will produce AI System Risk File 'Evidence Packages.' These are standardized templates that technology providers and utilities will use to document and prove the safety and reliability of an AI system. This creates a common language for risk, allowing a utility in California to evaluate a new AI-powered forecasting tool using the same rigorous criteria as a utility in New York.

Finally, the centerpiece will be an Open-Source Assessment Toolkit. This software will provide a practical, registry-backed workflow for tiered risk assessment. It will include testing protocols, verification and validation (TEVV) procedures, and monitoring runbooks to help organizations evaluate an AI system before it goes live and monitor it throughout its lifecycle. By making the toolkit open-source, EPRI is ensuring broad accessibility and encouraging continuous improvement from the wider community. This practical approach is designed to operationalize existing high-level frameworks like the NIST AI Risk Management Framework and NERC's Critical Infrastructure Protection (CIP) standards, translating them into actionable steps for the energy sector.

Taming the AI Paradox

The urgency behind SAFERai.power is rooted in the dual role AI is playing in the energy transition. On one hand, research from EPRI and others projects that electricity demand from U.S. data centers could swell from 4-5% of the national total today to as much as 17% by 2030. This exponential growth places enormous strain on an already complex grid.

On the other hand, AI is widely seen as the most powerful tool available to manage that complexity. AI-driven applications are already being deployed for predictive maintenance, wildfire detection, DER integration, and cybersecurity. The challenge, as one utility executive noted, is moving these applications from small-scale pilots to safely governed, enterprise-wide deployments.

This is where the work of the broader OPAI consortium, which SAFERai.power builds upon, becomes critical. Launched in March 2025, OPAI focuses on addressing the foundational challenge of "data readiness," creating open AI models and sandbox environments where utilities can test solutions without risking the live grid.

Reliability organizations are keenly aware of the stakes. “Understanding the potential applications for AI can help us understand both the benefits and work with industry to mitigate any risks,” said Jim Robb, president and CEO of NERC. “This effort will lay the foundation for that work.”

The sentiment is echoed by major utilities on the front lines of this transformation. “The utility sector is unique in that it requires a heightened bar for responsible and secure AI use to ensure the ability to provide reliable, resilient, and affordable power is enhanced and not compromised,” explained Robert Piascik, senior vice president and chief information officer at the New York Power Authority (NYPA). He emphasized that as AI accelerates, utilities need clear, evidence-based methods to evaluate risk and maintain public trust, which is precisely what the initiative aims to provide.

By bringing together the architects of AI, the operators of the grid, and the guardians of public trust, SAFERai.power is tackling one of the most complex and consequential challenges of the modern economy. The goal is not to slow down innovation, but to channel it responsibly, ensuring the AI-powered grid of the future is not only smarter, but safer and more reliable for everyone.