Gartner misconfigured AI critical infrastructure prediction warns of shutdown risk

A new forecast highlights a different kind of national risk. Not cyberattacks. Not natural disasters. But misconfigured artificial intelligence.

Gartner predicts that by 2028, misconfigured AI in cyber physical systems could shut down critical national infrastructure in a G20 country. The warning shifts attention from external threats to internal system design, configuration and governance failures.

What is at stake: Cyber physical systems

Gartner defines cyber physical systems, or CPS, as engineered systems that coordinate sensing, computation, control, networking and analytics to interact with the physical world, including humans.

CPS serves as an umbrella term covering:

• Operational technology

• Industrial control systems

• Industrial automation and control systems

• Industrial Internet of Things

• Robots and drones

• Industrie 4.0 environments

These systems underpin power grids, manufacturing plants and other essential infrastructure. Increasingly, they rely on AI models to automate and optimise decision-making.

The forecast suggests that the risk now lies not just in malicious intrusion but in how these AI systems are configured and maintained.

Misconfiguration: The overlooked vulnerability

According to Gartner, misconfigured AI can autonomously shut down services, misinterpret sensor data or trigger unsafe actions. The consequences may include physical damage and large-scale service disruption.

The report outlines a plausible scenario in modern power networks. AI models are used for real-time balancing of electricity generation and consumption. If a predictive model misreads demand signals as system instability, it could trigger unnecessary grid isolation or load shedding across regions or even entire countries.

Wam Voster, VP Analyst at Gartner, said the next major infrastructure failure may not be caused by hackers or natural disasters but by a well-intentioned engineer, a flawed update script or a misplaced decimal. He emphasised the need for a secure kill-switch or override mode accessible only to authorised operators to prevent unintended shutdowns.

The issue is compounded by the growing complexity of AI systems.

“Modern AI models are so complex they often resemble black boxes,” Voster said. He noted that even developers may struggle to predict how small configuration changes can affect system behaviour. As systems become more opaque, the risk posed by misconfiguration increases, reinforcing the need for human intervention mechanisms.

Governance, override and resilience

Gartner’s recommendation is clear: human control must remain central, even in highly autonomous environments.

To mitigate risks, the firm advises chief information security officers to implement several measures.

Safe override modes

All critical infrastructure CPS should include secure override mechanisms, such as a kill-switch, accessible only to authorised personnel. This ensures that human operators retain ultimate control during autonomous operations.

Digital twins for testing

Organisations should develop full-scale digital twins of infrastructure systems. These replicas allow realistic testing of AI updates and configuration changes before live deployment, reducing the likelihood of unintended consequences.

Real-time monitoring and rollback

Gartner also recommends continuous monitoring of AI changes within CPS environments. Systems should include rollback mechanisms to reverse problematic updates quickly. In addition, the firm suggests the creation of national AI incident response teams to manage and contain emerging risks.

A shift in the risk narrative

The Gartner misconfigured AI critical infrastructure prediction reframes how governments and enterprises view AI risk.

The threat is no longer limited to external adversaries. It includes internal complexity, configuration management and system opacity.

As AI systems take on greater operational responsibility in power grids, manufacturing and other essential services, the margin for error narrows. A small misstep in configuration could cascade into widespread disruption.

The forecast does not suggest inevitability. It highlights vulnerability.

For technology leaders and policymakers, the message is direct: autonomy without control is a systemic risk. Governance, transparency and override capability must evolve at the same pace as AI adoption in critical infrastructure.

Read More:

NetSuite AI-powered innovations reshape ERP workflows

AMD 41.3% server revenue share surges

Okta launches Agent Discovery in ISPM to combat shadow AI risks

Kaspersky study highlights SOC outsourcing in India