Benefit of AI in EHS Needs to Be at Enterprise Level
A recent survey, EHS: From Curiosity to Confidence, from EY, found that many companies are, in fact, using AI within the EHS function to improve efficiency. However, this is mostly limited to individual applications of general AI Chatbots. What is needed to improve EHS functions is shifting the technology to enterprise-level adoption.
“EHS performance has not been improving at the expected pace for some time now," said Patricio Estevez partner, Environment, Health and Safety, Ernst & Young, Australia, in the report. "While it’s certainly not a silver bullet, AI offers a tangible way for us to shift the dial - better protecting our environment and sending more people home safely every day.”
A practical pathway for AI-enabled EHS improvement across the enterprise
Instead of the typical adoption of AI through a technology-driven path, such as machine learning to deep learning, followed by generative AI and agentic AI, EY suggests focusing on AI from the type of technology to its purpose.
To help determine that EY has suggested using these four principles (excerpted below).
Intention: AI-enabled EHS improvement must begin with a clearly defined purpose. Rather than treating AI as a generic capability or technology upgrade, organizations should explicitly articulate why AI is being used, what EHS problems it is intended to address and what outcomes it is expected to support. This intent should be aligned with the organization’s broader strategic objectives, operating context and risk profile, as well as current and emerging EHS and organizational risks.
It is important to regularly measure and monitor whether the intended purpose of AI is being achieved, and to refine its application as needed to ensure it remains effective and aligned with organizational intent.
Integration: AI only strengthens EHS governance and performance when it is embedded into operating models, decision ownership, data and reporting, and assurance mechanisms. Integration with other operational functions is also critical, considering when data and/or processes could overlap and how to effectively embed this into the design and governance. As with all EHS systems and processes, the needs and capabilities of end users must be considered for the AI to be functional and effective.
Capability: Accurate data, clear ownership, and workforce trust are enabling conditions. Sustained AI adoption in EHS depends on clear accountability across EHS, technology, and risk functions, data that can be trusted and integrated into workflows, and early, transparent engagement with the workforce. While AI can be used to help provide structure to unstructured data, if its outputs are found to be incorrect or unhelpful, this presents an opportunity to revisit the data source(s) and adjust.
Human judgement: AI can inform decisions, but accountability for any decisions relating to EHS and associated outcomes must remain clearly owned by leaders. In safety-critical environments, AI should supplement professional judgement by frontline teams, not replace it. It is important, particularly when it comes to EHS, that AI on its own is not used to guide process flow or decision-making.
The report also offers some specific examples of how companies are using the technology.
How has AI been used to understand EHS risk?
Aviation, construction, and industrial safety have used natural language processing (NLP) to classify EHS incidents, extract causal mechanisms, and enable thematic risk analysis at scale.
AI is being used to continuously integrate and analyze real-time data from sensors, equipment logs, and operational systems to inform safety monitoring dashboards.
How has AI been used to act on EHS risk?
AI analyses real-time air quality data and generates actionable recommendations, such as adjusting traffic flow, modifying industrial operations, or issuing public health alerts when pollution levels exceed safe limits. This enables faster interventions that reduce EHS risk.
AI-driven computer vision systems have been used for real-time hazard detection linked to automated plant responses, including isolation of machinery when unsafe human–machine proximity is detected. In these situations, AI has been a substitute for manual monitoring and PPE-dependent controls.
AI-powered drones and robotic systems have been used in confined-space and high-hazard industrial environments specifically to remove workers from exposure. This limits reliance on PPE and procedural controls for hazardous access.
How has AI been used to anticipate EHS harm?
AI analyses large volumes of seismic, satellite, and ground-sensor data to help predict geological hazards (e.g., earthquakes, landslides, volcanic activity), enabling risk mitigation before events occur.
In advanced manufacturing, AI is used to predict tool failure before it occurs. Machine-learning models analyze real-time data from force, vibration and acoustic sensors to predict cutting tool wear. While this AI has been designed primarily to improve operational efficiency, it also enables tools to be replaced before wear leads to unsafe operating conditions.
AI identifies illegal waste dumping before it causes EHS impacts. AI systems analyze images from drones or surveillance cameras using deep-learning models to locate illegal dumping, enabling greater efficiency over manual methods and allowing authorities to intervene early and prevent soil, water, and ecosystem contamination.