The Trust Gap: Why AI Adoption Has Outpaced AI Readiness in EHS

TL;DR 

Ninety-four percent of EHS and sustainability departments across manufacturers, retailers, pharma, and chemical companies now use AI in some form, but only 9% have scaled it into governed, enterprise-wide decision-making. The gap is not driven by a technology problem. It is a data and domain expertise problem. In regulated environments where a wrong answer carries compliance, legal, and safety risk, responsible AI adoption starts with the quality of the knowledge and agentic reasoning powering the AI model, not simply the AI technology itself. 

AI Adoption in EHS has surged, but maturity has not 

In 2019, just 5% of environmental, health, and safety organizations reported using artificial intelligence. By 2026, that number had climbed to 94%, according to the NAEM State of AI in EHS and Sustainability report. 

On the surface, that trajectory looks like a success story with an impressive impact. But the maturity data on how AI is being used tells a very different tale and highlights the nuanced complexity of adopting AI in mission-critical workflows. 

Only 9% of organizations surveyed are using AI in ways that are scaled, governed, and trusted for enterprise-wide decisions. That means that the vast majority are still in early trial and error stages marked by experimenting with general-purpose tools, running isolated pilots, or applying AI to low-risk tasks that sit well outside core compliance workflows. 

This is the AI trust gap in EHS&S. Adoption has moved fast. Workflow impact has not. And the gap matters, because EHS is not a domain where close enough is acceptable. 

What the data reveals about where organizations are 

The NAEM report asked EHS professionals across a broad range of industries, organization sizes, and risk profiles on how they are using AI today. The picture that emerges is one of genuine progress at the individual level, and persistent caution at the organizational one. 

The most common AI applications in EHS today are practical and productivity-focused: searching for and synthesizing information (94%), summarizing regulations and guidance documents (81%), and analyzing data for key takeaways (72%). These are real efficiency gains, and they reflect what AI does well when deployed for general research and content tasks.   

But the least common applications tell a more significant story. Automated compliance monitoring across jurisdictions, predictive models for environmental incidents, and AI-supported regulatory submissions each appear in fewer than 5% of organizations. These are precisely the workflows where AI could have the greatest impact on compliance outcomes, and they're also the ones that demand the most from the systems and data behind them.  

These are general-purpose productivity gains, but EHS is not a generic AI use case, and the survey data bears that out. 

Why data and domain expertise matter more than ever 

The NAEM findings align with what practitioners in regulated industries already know from experience: the quality of an AI output is bounded by the quality of the data it draws from.  

In EHS, that data is not generic web content or unstructured internal documents. It is curated, jurisdiction-specific, regulatory intelligence, maintained by domain experts who understand the underlying science, the regulatory landscape, and the operational context in which compliance decisions are made. An AI system answering a question about chemical classification needs to account for differences in GHS implementation across dozens of countries. A tool supporting SDS authoring needs to interpret substance-specific regulatory thresholds that shift as new hazard data emerges. A compliance monitoring application needs to track regulatory changes across multiple jurisdictions in near-real time.  

This is where the concept of domain expertise in AI becomes concrete rather than abstract. It is not just a matter of which model an organization uses. It is the depth of the authoritative data behind it; the kind of structured, expert-maintained knowledge base that AI systems need to produce outputs EHS professionals can trust and defend. 

What responsible AI readiness looks like in practice 

Responsible AI readiness in EHS starts with a foundation of trustworthy data. The organizations best positioned to get there are the ones whose AI is grounded in authoritative regulatory intelligence, not retrieved from the open web. Open LLMs often rely on broad, web-sourced training data, which can introduce variability in answer quality. A SEMrush analysis of AI-generated responses found that Reddit was the most frequently cited source, followed by YouTube, highlighting the prominence of community-generated content in AI outputs.  

At 3E, that foundation is the product of 35 years of work: more than 500,000 substances mapped to regulatory requirements, coverage of more than 3,000 regulatory topics across 160+ countries, and continuous monitoring of more than 500 regulatory lists, maintained by a team of domain experts who track, interpret, and structure that intelligence in real time.  

What this looks like in practice is an AI platform where that foundation powers multiple purpose-built products, each applying agentic reasoning to a specific compliance workflow. In 3E Insight, the AI Assistant delivers fast, traceable answers to complex regulatory questions, drawing on full-text regulations, curated substance lists, and global regulatory news across 160+ countries. In 3E Protect, the Portfolio Assistant screens a company's full chemical inventory against global thresholds, surfaces the highest-risk items first, and provides citations and a verifiable audit trail for every answer. And the 3E Regulatory Agent automates horizon scanning and impact analysis workflows, connecting regulatory changes directly to the substances and markets they affect so compliance teams can act before a deadline rather than after. 

3E's agentic reasoning models apply that knowledge base across multiple AI use cases in a single platform. In regulatory intelligence, AI agents monitor rule changes across hundreds of jurisdictions, flag substance-level impacts, and surface compliance gaps before they become audit findings. In product stewardship workflows, AI supports chemical inventory screening, SDS authoring, and downstream customer notification. In supply chain compliance, AI agents help teams assess supplier documentation, identify regulatory exposure across product portfolios, and track obligation changes that affect sourcing decisions. Each use case runs on the same governed data foundation, which means outputs are traceable, defensible, and consistent across the platform. 

This architecture, expert-maintained data combined with agentic reasoning purpose-built for regulated work, is documented in 3E's AI Trust Center, which sets out how data privacy, model transparency, and responsible AI practices are embedded into product development. 

Questions every EHS leader should ask before scaling AI 

The barriers the NAEM data surfaces point to a set of questions worth asking before scaling AI in your compliance and safety workflows: 

• What data is the AI drawing from? Is it trained on general web content, or does it use curated, regulatory-grade intelligence maintained by domain experts? 

• How does it handle jurisdictional complexity? Can the system account for regulatory differences across countries, regions, and substance categories? 

• What governance is in place? Are there documented practices around data privacy, model transparency, and auditability? 

• Where is the human in the loop? Does the workflow include expert review at critical decision points, or does it rely on AI outputs without validation? 

• Can you defend the output? If a regulator questions a compliance decision made with AI support, can you trace the answer back to an authoritative source? 

These are not theoretical questions. They are the practical criteria that the NAEM data shows most organizations are still working to answer, and that distinguish organizations experimenting with AI from those ready to trust it at enterprise scale. 

The full findings from the NAEM State of AI in EHS and Sustainability report, including maturity benchmarks, barrier analysis, and peer insights, are available here: 

Download the Report.