The capability–governance gap: A 2026 evidence review of public LLMs in regulated work

Public large language models like ChatGPT, Gemini, and Claude are transforming regulated industries — but are they safe enough for compliance-critical work?

The Capability–Governance Gap is a research-backed report examining the latest 2025–2026 evidence on how public LLMs perform across regulatory science, chemistry, pharmacovigilance, EHS, product stewardship, and compliance operations. Drawing on peer-reviewed studies, the report explores where AI systems deliver strong results — and where they still fail in ways that create operational, legal, and safety risk.

The findings challenge two common assumptions:

• that public LLMs are too unreliable for regulated environments, and
• that enterprise AI alone is sufficient for production-safe compliance workflows.

Instead, the evidence points to a more practical conclusion: AI capability is no longer the primary limitation — governance architecture is.

Inside the report, readers will discover:

• Where public LLMs already achieve expert-level accuracy in regulatory and chemistry workflows
• Why hallucinations, citation failures, prompt injection, and weak hazard reasoning still create serious deployment risks
• The six governance-layer requirements for production-safe AI in regulated operations
• How organizations can combine AI scale with human expertise for defensible compliance outcomes
• Why hybrid architectures are emerging as the safest and most scalable approach to AI-enabled regulatory work

Whether you lead product compliance, EHS, sustainability, regulatory affairs, chemical management, or digital transformation initiatives, this report provides a grounded, evidence-based framework for evaluating AI in high-stakes environments.

Download the report to learn how leading organizations can responsibly deploy AI while maintaining traceability, accountability, and regulatory confidence at scale.