AVEVA Outlines Guardrails for Autonomous Industrial AI

Executive Summary

• AVEVA chief technologist Arti Garg argues that industrial AI is entering a new phase, driven by foundation models, physical AI and agentic AI that can automate more complex tasks across industrial environments, according to MIT Tech Review AI.
• Unlike purely digital AI, industrial AI interacts directly with physical systems, where an unexpected decision can carry consequences for safety, reliability and critical infrastructure, the report states.
• AVEVA frames responsible AI around a triple mandate: secure, efficient (including environmentally efficient), and preserving human safety and oversight, with AI augmenting rather than replacing people in critical decision loops.
• Garg chairs an IEEE working group, the P7100 Standards Working Group, developing a standard methodology for measuring AI's environmental impact across electricity, energy, resources, water and carbon.
• The report, produced in partnership with AVEVA, identifies data correlation across telemetry, service logs and engineering documents, workforce retirement and grid management as central operational pressures.

Key Takeaways

• Industrial AI adoption is accelerating, and the report cites one study suggesting roughly a 78% increase over the past two years within the industrial sector.
• The core governance question, per Garg, is how to leverage less predictable models while maintaining safety and reliable operations.
• AVEVA's stated approach favors human-in-the-loop recommendations over unchecked closed-loop automation, with guardrails limiting where automated systems can act.
• Garg says almost half of the industrial workforce is set to retire in the next five years, creating both a knowledge-loss risk and an opportunity to capture expertise in AI systems.

MIT Tech Review AI Examines Industrial AI's Shift Toward Physical Systems

The central argument in the MIT Tech Review AI report is that industrial AI is not new — AVEVA says it has worked on AI applications in the industrial sector for more than 20 years — but the type of AI has changed. General-purpose foundation models, physical AI that accelerates robotics and autonomous systems, and agentic AI that automates at the software level have widened who can deploy advanced AI. The report describes this as an inflection point that behaves almost like a step function in adoption.

What makes the shift consequential is the operating environment. Industrial AI touches pumps, mixers, power systems and mining equipment, and often sits in hazardous settings. Garg's framing is that the biggest risk is the interaction with real physical systems capable of delivering outcomes that keep the world running. The report also notes that newer models are not fundamentally explainable in the way prior statistical or anomaly-detection models were, and that their behavior can change as they learn and are tuned.

MIT Tech Review AI Details AVEVA's Data Correlation Problem

The report places data at the foundation of the transition. Industrial systems hold information across telemetry, service logs, engineering documentation and other disparate sources. AVEVA's description of the problem is concrete: correlating readings from a pump with the last maintenance log and with original design documentation so an operator can diagnose a fault more quickly.

Garg says newer technologies, including graph databases and AI matching of disparate data sets, shorten that research. The report describes an operator using an iPad with an AI that fetches and correlates information in near-real time. It extends the example to robots gathering data in hazardous environments — either with onboard computation or a quick connection to an operator who no longer has to enter a dangerous space.

MIT Tech Review AI Reports on Governance, Guardrails and Human Oversight

AVEVA's responsible AI framework, as described in the report, emphasizes security, efficiency and human safety and oversight above all. The company describes a multi-layered governance approach spanning how it uses AI internally and how it deploys AI into products.

On autonomy, the report lays out a specific progression: from human operator to human supervisor of systems. In a human-in-the-loop arrangement, a system processes data and makes a recommendation. AVEVA says it has worked with customers to ingest operational data and simulations and recommend operating set points. There is interest in moving from recommended set points to automation that adjusts them automatically, and the report says AVEVA has had successful real-world pilots around that. Garg acknowledges this makes people nervous, which is where guardrails — such as limits on operating bands or restricting automation to certain areas — come in.

Related: Graftcode Raises €2.1M for Software Integration Platform in 2026

Garg is candid that the sector is still working out the balance. Her point is that AI systems are not human beings, so the guardrails look different, and the appropriate design likely depends on the application.

MIT Tech Review AI on Sustainability, Standards and Grid Resilience

The report connects sustainability to industrial AI in two directions. First, AI can help manage complex power systems as renewable generation grows. Garg cites AVEVA's partnership with Idaho National Laboratory as part of its testing for an AI grid resilience project, focused on helping grid operators understand system conditions and detect anomalous behavior early. The challenge she describes is managing a grid with many smaller, intermittent and distributed sources rather than one spinning asset delivering frequency.

Second, the report raises the environmental footprint of AI itself. Garg says there is no agreed-upon method to measure AI's environmental impact, despite widely circulated figures about water and electricity use. She chairs the IEEE P7100 Standards Working Group, launched a little over two years ago, which aims to define one methodology covering electricity and energy consumption, resource usage, water consumption and carbon, usable for reporting and potentially for oversight or regulatory purposes. She also argues that choosing the right size model for the right application is itself a path to greater environmental efficiency, with the ancillary benefit that purpose-built models are less likely to misbehave unpredictably.

For deeper context, see our Robotics analysis: "Siemens, ABB and Honeywell Push Enterprise Robotics Integration".

MIT Tech Review AI Signals

No other entity, geography or date is supported by the supplied source, and none has been added.

MIT Tech Review AI Implementation Risks

The report identifies several risks that organizations should weigh before expanding autonomy. Physical interaction is the first: industrial AI can act on systems with real-world safety, reliability and critical-infrastructure consequences. Model opacity is the second: the newer capabilities are not fundamentally explainable in the way prior statistical models were, and their behavior can change as they learn and are tuned. Governance is the third: Garg says the sector is still figuring out where automated systems may act and where human supervisors remain responsible, and that guardrails for AI look different from those for human staff. Workforce transition is the fourth: the report cites a large share of the industrial workforce due to retire within five years, which places pressure on capturing expertise before it leaves. Measurement is the fifth: there is no agreed method for the environmental impact of AI, which complicates both reporting and oversight.

Editorial independence disclosure: this article analyzes a report produced by MIT Technology Review Insights in partnership with AVEVA, a commercial relationship that readers should weigh when assessing its claims.

Additional coverage: Klarna vs Intercom Fin vs Sierra: Conversational AI Compared for 2026

Source note: all factual details above derive from MIT Technology Review's report on building a safer path to autonomous industrial AI. No additional sources were verified for this article.

What This Means for Practitioners

For CIOs, plant operators and procurement teams, the report's practical message is that autonomy should be staged, not switched on. AVEVA's path runs from AI-assisted diagnosis of correlated telemetry, logs and engineering documents, to recommended set points, to pilots where set points adjust automatically within defined bands. Teams evaluating vendors should therefore ask where human oversight sits, what the guardrails actually restrict, and how model behavior is monitored as it changes. The workforce data point matters too: with a large share of experienced operators nearing retirement, capturing that expertise in systems is both an operational and a knowledge-retention decision.