Industry executives and experts share their predictions for 2026. Read them in this 18th annual VMblog.com series exclusive.
By Aron England, CPTO at Accruent
Enterprise AI is quickly moving from a productivity aid to an operational decision-maker across industries like manufacturing, healthcare, and energy. However, as this shift gains momentum, a long-standing technical issue is becoming a serious business and safety concern: AI hallucinations.
In these high-stakes environments, made-up outputs aren’t only inaccurate; they can undermine compliance, introduce physical risk, and expose organizations to regulatory consequences. What was once tolerable in low-risk use cases, will become increasingly dangerous when AI recommendations are acted on in the real world.
As we move into 2026, vendors will need to rethink how they deploy AI in regulated areas, prioritizing transparency, traceability, and accountability before hallucinations turn into irreversible damage.
Rising AI Hallucinations & Compliance Risks
As AI becomes relied on for faster decision-making, these models are also being pushed beyond simple retrieval into interpretation. Large Language Models (LLMs) often create answers from multiple pages without citing back to the original evidence. Especially as the volume of documents and versions grows within a model, so does the risk of hallucinated output, introducing significant challenges.
For one, these hallucinations cause major compliance risks for frontline workers relying on AI-generated recommendations when working on equipment and machinery. For example, if a technician follows an AI-generated step that isn’t grounded directly in an OEM manual, the worker may fail to resolve a machinery issue-or worse, the organization could face serious compliance risk.
When something goes wrong or the issue wasn’t solved, following an AI hallucinated recommendation also means there may be no auditable evidence to explain why that specific action was taken, leaving teams on their heels during safety reviews or regulatory investigations.
In addition to traceability problems, enterprise AI tools will increasingly ask users to “trust the model” without explaining exactly how conclusions were reached. If a worker asks AI to analyze 900 leases, they should know whether it evaluated all 900 or only 300, and why or why not. If this lack of transparency is not addressed, it will only increase the likelihood of hallucinations, or even omit critical information.
To close the growing trust gap in today’s enterprise AI, it’s vital in the next few years that tools can trace back to the original content used, and can also highlight which pages or documents were actually evaluated so users know whether the AI reviewed all relevant documents or only some.
Top Safety Risks in High-Stake Industries
AI hallucinations are becoming more than an inconvenience in regulated industries. They will create real safety and operational risks.
For example, in a manufacturing setting, if an AI-generated suggestion tells a worker to proceed with a repair or take a wrong action on a piece of equipment, it could cause unplanned downtime, damaged machinery, wasted material, or defective end products that cause recalls, deteriorating customer trust and loyalty.
Healthcare is another industry where these AI hallucinations can cause massive risks. If a machine failure at a hospital is not properly fixed by a technician following AI-generated recommendations, or not fixed in time, patients’ lives are at stake.
For these mission-critical roles in industries tied to physical operations, mistakes carry consequences that can’t be erased. This is why AI systems operating in critical environments should be designed to explain, justify, and withstand scrutiny, not just produce answers.
The Importance of Transparent, Audit-Ready AI in 2026
In the next year, AI recommendations should be able to be traced back to meaningful output points in its source material, such as the specific manual page, diagram, data table, or log entry that informed the suggestion. This starts with strategically capturing manuals, drawings, leases, and historical work orders in one place, to ensure AI is providing answers directly from a company’s arsenal of information, not generic training data.
Without using tools to ensure discoverability, technicians who are already under intense pressure will have to spend even more time searching documents manually to verify accuracy.
Overall, the fundamental issue that companies will need to address in the next year is transparency. Trust in enterprise AI will increasingly depend on vendors that can provide explicit details about data usage, such as what content the AI accessed, ignored, how data is being utilized, and if it feeds back into training.
In the next couple of years, vendors that clearly state AI limitations, keep humans in the loop for crucial decisions, and start with well-bounded use cases that deliver value without turning critical workflows into a black box, will earn the best credibility in high-stakes industries.
##
ABOUT THE AUTHOR
As Chief Product and Technology Officer, Aron England leads product strategy and drives technology innovation, accelerating the delivery of agile, customer-centric solutions. With over 25 years of experience, Aron is a seasoned product leader who has successfully scaled B2B and B2C SaaS and agentic AI solutions from research through successful high-growing customer-facing products. He has built and led high-performing global product teams at Rev, Dell, Vrbo and Expedia. Aron holds a bachelor’s degree from Western Washington University in Bellingham, Washington.
