Why “80% accurate” is the biggest risk in AI compliance models right now

Perspectives from 3E's 2026 customer conferences by Greg Gartland, CEO, and Alan Johnson, Managing Director, Chemical & Workplace Safety, 3E -Spring 2026

Key takeaways

• The “80% problem” is real. AI that gets you 80% of the way there still leaves you 100% responsible for the remaining 20%-and in regulated environments, that’s where risk, including violations and liability, lives.
• Three things make AI trustworthy for compliance: domain expertise, proprietary data, and a reasoning layer built for governance and traceability. Leave one out, and risk can increase exponentially.
• Auditability is not negotiable. Every AI output in a compliance workflow should be traceable to source data, repeatable, and defensible to a regulator-not a black box.
• Agentic AI raises the stakes. As AI moves from answering questions to executing workflows, the consequences of unreliable data shift from wrong answers to wrong outcomes.
• Ask every AI vendor five questions. Are outputs auditable? Reproducible? Built on expert-curated data? Backed by a published trust framework? Stress-tested against your actual regulatory workflows?

Every environment, health, and safety (EHS) leader, product steward, and compliance professional in our industry is hearing the same message right now: adopt AI. The pressure is real. Product regulatory compliance, particularly in the area of chemicals and hazards regulation, is accelerating faster than teams can absorb. Headcount is flat. Obligations are not.

We get it. We're regulatory experts and technologists. We've been building AI, machine learning, and natural language processing capabilities at 3E for over a decade-long before generative AI entered the conversation. In fact, we're literally “all in on AI”, announcing our new 3E AI Platform earlier this year to prove it.

But here's what concerns us: the conversation about AI in regulatory compliance has become almost entirely about capability. What can it do? How fast can it go? What we rarely hear anyone ask is the question that really matters in regulated environments: Can you stand behind the outcomes?

The 80% problem: Why nearly-right AI is a category of risk

At our customer meetings across the US, Europe, and Japan, known as “Engage26”, we shared a story that crystallized this risk. We heard about a vendor in our space that built an AI capability and described it as getting customers “about 80% of the way there.”

Eighty percent sounds impressive. But think about what that means for chemical compliance and workplace safety. If a tool gets you 80% of the way there, you're still 100% responsible for the remaining 20%. In product compliance, safety data sheet management, and chemical regulatory workflows, 20% isn't a rounding error. It's where regulatory violations live, exposure hides, and the defensibility of your decisions breaks down.

Nearly-right AI is a distinct category of risk in this industry. Generic large language models produce answers that sound credible-and often are-but in the product stewardship and regulatory intelligence workflows you're responsible for, “often” isn't a standard you can defend. When a regulator asks how you arrived at a REACH, TSCA, or Proposition 65 classification, or why a substance was cleared for a specific use, “the AI said so” isn't an answer. You need to know how the AI arrived at that conclusion, what data it used, and whether you can reproduce the result.

This is especially urgent right now. NAEM's 2026 “State of AI in EHS and Sustainability” research found that while organizations are increasingly piloting AI in compliance workflows, very few have moved to production-level deployment, precisely because the trust gap has not been closed.

The 3E AI Platform is a trust by design artificial intelligence platform that combines 35+ years of proprietary regulatory data, 160-plus domain experts, and an 81-step quality assurance process to deliver auditable, traceable AI outcomes for chemical regulatory compliance, product stewardship, and supply chain transparency workflows.

What trustworthy AI for compliance actually requires

We believe trustworthy AI compliance solutions aren't just built on a better model. They are built on a core formula: world-class domain expertise + proprietary regulatory data + an AI reasoning layer engineered for governance and traceability. Remove any one of those elements and the system breaks down.

Domain expertise means having 160-plus scientists, toxicologists, chemists, and regulatory specialists-many of them PhDs-who understand the intent behind regulations, not just the text. These experts have spent 35 years curating, interpreting, and validating the chemical regulatory intelligence that powers every 3E product. Trusted by EHS teams across chemical manufacturing, pharmaceuticals, industrial distribution, and other verticals, that human expertise is the quality layer that makes AI outputs defensible.

Proprietary regulatory data means a continuously curated, tagged, and validated data foundation, not the open web. Every safety data sheet in our SDS management system-the world's largest safety data sheet library, with over 20 million SDS documents-is verified at least every 12 months for accuracy, recency, and completeness. We screen against more than 200 global regulatory frameworks. Every news article from our regulatory horizon scanning solution is tagged with CAS numbers, regulatory metadata, and jurisdictional information. We're so confident in this data that we now publicly guarantee it, and back that guarantee with a service commitment.

An AI reasoning layer built for governance means specialized AI scaffolding to ensure security, versioning, and plug and play model iteration, a multi agent system capable of expert orchestration, an 81-step quality assurance process to ensure trust by design, and a launch partner program where we actively engage customers on the technology, listen to early feedback, and see concrete user adoption before any AI capability ships It means outputs that are auditable, traceable, and repeatable. It means evidence packets that travel with every AI agent and model context protocol (MCP) connector we deploy, so your IT and CISO teams understand exactly how the AI reasons, what data it accesses, and what safeguards are in place.

Explore how the 3E AI Platform delivers this trust-first approach across your compliance workflows.

What trusted AI looks like in practice

The trust formula matters most when the regulatory stakes are highest. Alan Johnson, our Managing Director of Chemical & Workplace Safety, works at that intersection every day. Here’s how he framed it at Engage26.

“Consider a real scenario. New Mexico passes new PFAS regulations. You manage a portfolio of 100,000 safety data sheets across multiple facility locations. How do you know which materials in which facilities are affected?

Without trusted AI, this is weeks of manual cross-referencing-the kind of compliance gap that keeps EHS teams up at night. With the kind of portfolio-level reasoning our AI capabilities in 3E Protect enable, you can surface relevant exposures across your entire chemical inventory-cross-referencing the new regulation against your facility tree, your SDS library, and your substance data-in a fundamentally different timeframe. And critically, you can trace exactly how the AI arrived at each result.

The same logic applies across product compliance and supply chain solutions. Through 3E Exchange, AI-powered supplier data collection helps suppliers answer hundreds of compliance questions using existing environmental product declarations and SDS-turning an exponential supply chain transparency problem into a manageable workflow. Through 3E Insight, the embedded AI assistant already helps hundreds of customers interrogate regulatory monitoring data, schedule personalized updates, and surface the changes most relevant to their specific substance portfolios,” - Alan Johnson, Managing Director, Chemical & Workplace Safety, 3E

That kind of traceability-from regulatory trigger to facility-level exposure to defensible outcome-is exactly the standard we think every AI vendor in this space should be held to. Which brings us to the practical question of how to evaluate them.

Beyond chat: AI agents that execute compliance workflows

The industry has largely moved past the “chat with documents” phase of AI adoption. The next frontier is agentic AI-capabilities that move beyond question-and-answer into executing multi-step compliance workflows.

“Compliance automation tools like our portfolio agent in 3E Protect, which can reason across large chemical inventories to surface regulatory exposures automatically. It means a chemical response agent that fast-forwards incident management reporting-from spill documentation to EPA notifications. And it means AI-powered SDS authoring automation and compliance letter generation that dramatically reduce turnaround time-turning what once required days of manual safety data sheet authoring into a streamlined, expert-validated workflow.” - Greg Gartland, CEO, 3E

But agentic AI amplifies the stakes. An agent operating on unreliable data doesn't just give you a wrong answer-it gives you a wrong outcome. That's why the trust infrastructure matters more, not less, as AI-driven compliance monitoring becomes more powerful.

A standard for evaluating AI in compliance

We're sharing this perspective as a practical standard-one we would apply to any AI vendor in this space, including ourselves. Every compliance and EHS team evaluating AI right now needs a framework for separating tools that are ready for regulated workflows from tools that are not.

Ask five questions of any vendor:

1. Are the AI outputs auditable and traceable to source data?
2. Can the AI reproduce the same result given the same inputs?
3. Is the underlying data curated by domain experts, or scraped from the open web?
4. Does the vendor publish a trust framework and governance documentation?
5. Has the AI been stress-tested against the specific chemical regulatory compliance workflows it claims to support?

We have published our own answers at the 3E Trust Center, where we detail our ISO 27001 framework alignment, SOC 2 Type 2 certification, and AI governance practices. We think that kind of transparency should be table stakes for any AI vendor operating in regulated environments.

The road ahead

The regulatory landscape is not slowing down. The 4,500-plus proprietary regulatory news articles we published last year reflect the pace of change that's accelerating-new substance restrictions, evolving supply chain transparency requirements, and regulations like the EU Sustainable Products Regulation creating exponential data demands that manual processes simply cannot absorb.

For EHS and product stewardship teams, that pressure is now existential. AI is the only viable path through it. But adoption without accountability is just a faster way to get the wrong answer. In regulated environments, the cost of a wrong answer isn't an audit finding; it's a worker exposed to an unidentified hazard, a product recalled after it reaches market, a supply chain violation that triggers regulatory action across multiple jurisdictions.

That’s the standard we're building to at 3E. Recognition from Verdantix as a leader in product compliance software and from Fast Company as one of the top 10 most innovative data science companies in the world is one signal. The real proof is in what our customers can defend-not 80% of the way, but all the way.

To learn more about 3E's approach to trusted AI for chemical regulatory compliance, visit the 3E Trust Center or explore the 3E AI Platform.