Positive AI in Energy Procurement: From Risk to Results

AI has a PR problem—partly because it’s earned it. Even many scientists who build and study these systems are more worried than excited. The loudest headlines are about deepfakes, misinformation, labor exploitation in data labeling, and the very real power grab of a few platforms owning too much of the AI stack.

Here’s my take: if you work in energy and utilities—especially in supply chain and procurement—pessimism is the most expensive stance you can take. Not because the risks aren’t real, but because your world runs on long-lived infrastructure, regulated accountability, and tight margins. If you sit out, you don’t stop AI. You just leave the choices to vendors and competitors.

The good news is that the research community is sketching a practical “positive vision” for AI: build norms, resist harmful use, apply AI responsibly to real problems, and strengthen institutions so they can govern AI. That blueprint maps cleanly onto energy procurement—where decisions about transformers, switchgear, grid-edge devices, DER interconnection hardware, and maintenance contracts directly shape reliability and decarbonization.

Scientists are worried—and that should sound familiar to utilities

A clear signal from the RSS article: optimism about AI isn’t evenly distributed. A Pew study (April) reported 56% of AI experts predict AI will have positive effects on society, but broader scientific communities show more concern than excitement (an Arizona State University-affiliated survey cited in the piece).

That split mirrors what I hear in utilities and critical infrastructure:

• Operations teams see opportunities (fewer outages, faster restoration, better planning).
• Risk, compliance, and cybersecurity see a widening attack surface.
• Procurement gets stuck in the middle—asked to move faster while being the last line of defense against vendor hype.

This matters because AI adoption in the grid supply chain is already happening—through vendor roadmaps, “AI-enabled” asset management platforms, forecasting tools, and customer engagement systems. Even if your organization isn’t “doing AI,” your suppliers probably are.

A utility that doesn’t shape AI requirements in procurement doesn’t avoid AI risk—it imports it.

A “positive vision” for AI in energy starts with problems we can measure

The article argues scientists need a positive vision, not just warnings. Energy and utilities have an advantage here: your success metrics are concrete—SAIDI/SAIFI, heat rate, losses, crew productivity, capex delivery, interconnection timelines, and safety incidents.

So what does “AI for public good” look like in this domain?

Grid optimization that doesn’t punish the most vulnerable

AI-assisted grid optimization can reduce losses, improve voltage management, and plan feeder upgrades—especially as EV charging and electrification loads grow. But “optimize” can become code for cost shifting.

A positive vision sets procurement requirements like:

• Equity constraints in optimization (don’t concentrate curtailment or outages in specific neighborhoods)
• Explainability for decisions affecting service quality
• Auditability so regulators can see what the model did and why

Renewable integration that rewards flexibility, not chaos

As utilities add renewables and connect more DERs, forecasting errors and congestion costs rise. AI can help predict load, solar output, and wind ramps, but only if the data pipelines are trustworthy.

Procurement’s role is unglamorous and essential: require data lineage, model monitoring, and fallback modes so grid operators aren’t left guessing when a model drifts.

Predictive maintenance that doesn’t become “predictive paperwork”

Predictive maintenance is one of the few AI use cases that can pay back fast—if it’s tied to work management reality.

The positive version isn’t “a model that predicts failures.” It’s:

• A model that triggers actionable work orders
• With spare parts forecasting tied to your supply chain constraints
• And measurable outcomes (reduced forced outages, fewer truck rolls, less overtime)

The 4 actions from science translate into a procurement playbook

The RSS piece frames four actions: reform the industry, resist harmful uses, use AI responsibly for good, and renovate institutions. For energy procurement and supply chain teams, that becomes a practical checklist.

1) Reform the vendor ecosystem with contract terms that force reality

Answer first: If you want ethical, sustainable AI, write it into sourcing and contracts.

Most companies get this wrong: they evaluate AI tools like ordinary software and then act surprised when models behave differently after deployment.

Put these into RFPs and MSAs for AI-enabled supply chain, asset management, and forecasting solutions:

• Training data disclosures: what data was used, what was excluded, and what licensing/permissions exist
• Model governance deliverables: model cards, risk assessments, and monitoring plans
• Performance definitions: specify what “accuracy” means (false negatives vs false positives) and the cost of each in your context
• Right to audit: technical and process audits, including subcontractors (data labelers, model hosting providers)
• Energy and carbon reporting: require reporting for model training and inference footprints for material workloads
• Vendor lock-in controls: portability of data, embeddings, and configuration; clear exit plans

For utilities, this also reduces regulatory risk. When a commission asks “how did you choose this system?”, a contract-backed governance trail is the difference between confidence and chaos.

Procurement example: critical spares and transformer lead times

Transformer constraints aren’t theoretical. Lead times and shortages can derail capital plans and storm response readiness.

AI demand forecasting can help—if it uses the right signals (asset health, failure probability, weather exposure, planned retirements) and if it’s integrated with supplier capacity realities.

A strong contract requirement here is decision traceability: when the model recommends stocking ten units of a specific class, you need to reproduce why. Otherwise, you can’t defend inventory strategy to finance, regulators, or auditors.

2) Resist harmful AI uses by documenting and blocking them early

Answer first: Your organization needs a short list of “no-go” AI patterns—before a vendor sells them to you.

The article emphasizes resisting harmful applications by documenting and shining a light on inappropriate use. In energy supply chain and procurement, harmful uses often show up as “efficiency features.”

Examples that deserve explicit guardrails:

• Opaque supplier scoring that penalizes smaller or minority-owned vendors because of biased historical data
• Automated contract risk scoring that isn’t explainable (and quietly nudges teams away from competition)
• Synthetic supplier references or AI-generated compliance artifacts (a real issue as AI-generated “slop” floods due diligence)
• Uncontrolled LLM tools summarizing contracts or creating SOWs without strict confidentiality and retention controls

A practical move: adopt a procurement policy that treats AI-generated artifacts as drafts, not evidence. Require human verification for certifications, safety documents, and supplier attestations.

3) Use AI responsibly where it improves reliability and decarbonization

Answer first: Responsible AI in energy procurement means picking use cases where humans stay in charge and outcomes are measurable.

The RSS article lists positive examples like translating under-resourced languages, scaling climate-change dialogue, and accelerating science via national labs. In utilities, the analog is scaling expertise—especially as retirements and workforce gaps hit protection engineering, substation design, and field operations.

Responsible, high-value use cases in the supply chain and procurement series context:

Supplier risk management that’s actually predictive

Instead of quarterly scorecards, AI can monitor leading indicators:

• delivery performance trends
• quality escapes and NCR patterns
• cyber posture changes
• financial stress signals
• geopolitical exposure by tier

But do it responsibly: don’t rely on black-box “risk scores.” Require features you can see (what drove the alert) and enforce a process where risk triggers a conversation, not an automatic disqualification.

Contract analytics that reduces disputes

LLMs can extract terms, compare redlines, and flag missing clauses. In utilities, the value is avoiding disputes that delay energization or maintenance work.

The responsible version includes:

• on-prem or controlled environments for sensitive contracts
• retention limits
• logging of prompts and outputs
• mandatory review for any clause suggestions

Forecasting that improves planning, not just dashboards

If you’re forecasting capex material needs, the model has to reflect realities like:

• long lead items (switchgear, breakers, relays)
• inspection hold points
• qualified supplier lists
• warehousing constraints

Otherwise AI becomes “predictive paperwork”—pretty charts that don’t change outcomes.

4) Renovate institutions: build AI governance that procurement can run

Answer first: AI governance fails when it’s only a policy document; it works when it’s a workflow.

The RSS piece calls for renovating institutions—universities, professional societies, democratic organizations. Utilities have their own institutions: asset management councils, model risk committees, cybersecurity programs, and procurement review boards.

What works in practice is a lightweight governance stack:

1. AI intake form (purpose, data sources, decision impact, human override)
2. Risk tiering (low/medium/high based on safety, reliability, customer impact)
3. Pre-deployment testing (bias checks for supplier scoring, stress tests for forecasts)
4. Operational monitoring (drift detection, incident process, rollback plans)
5. Post-incident learning loop (model and process updates, not blame)

Make procurement a co-owner. Procurement sees the vendor ecosystem earlier than anyone else and can stop bad patterns before they enter production.

People also ask: practical questions energy procurement teams raise

“Do we need our own models, or can we buy everything?”

Buying is fine for many functions, but critical decisions need controllability: clear data rights, auditability, and the ability to validate outputs. For outage risk, safety-related maintenance, and high-value sourcing, don’t accept “trust us.”

“How do we handle AI’s energy use when we’re trying to decarbonize?”

Treat AI compute like any other operational footprint:

• require reporting
• right-size models (smaller can be better)
• prioritize efficient inference
• avoid retraining unless it changes outcomes

The positive vision is simple: don’t burn megawatt-hours for marginal gains.

“What’s the fastest way to get value without creating a governance mess?”

Start with narrow, measurable supply chain AI: spare parts forecasting for a single asset class, supplier risk alerts for a few strategic vendors, or contract clause extraction with human review. Prove value, then scale with controls.

Where this fits in the AI in Supply Chain & Procurement series

This series is about demand forecasting, supplier management, risk reduction, and optimization. The missing ingredient is usually intent.

If your only goal is efficiency, you’ll get efficiency—plus fragility. A positive vision aims for reliability, fairness, auditability, and sustainability, because energy supply chains are public-interest supply chains. When they fail, communities pay.

The goal isn’t “more AI.” The goal is infrastructure we can trust.

Most organizations already have the raw materials to do this: sourcing discipline, safety culture, and regulated accountability. Add clear AI requirements, measurable outcomes, and governance that matches the risk.

If you’re planning 2026 initiatives right now, here are strong next steps:

• Choose one procurement-led AI pilot tied to reliability (not a generic chatbot)
• Update RFP templates with auditability, data rights, and monitoring requirements
• Create a “no-go list” for supplier scoring opacity and unverified AI-generated compliance artifacts
• Stand up an AI review workflow that procurement and operations share

The research community is right: we can’t build a positive future without a vision of what it looks like. For energy and utilities, that vision is pretty concrete—a grid that’s more reliable, faster to repair, cleaner to operate, and harder to game. The open question is whether your procurement team will write that vision into the contracts now—or inherit someone else’s later.