REGULATION & COMPLIANCE reviewed by people who understand the regulatory context.
Good AI use
Good AI use in chemical compliance starts before the prompt is written. The user must know the task. Is the goal to summarise, compare, translate, classify, draft, challenge assumptions, extract obligations or prepare questions for a supplier? These are different tasks and a vague prompt produces a vague result.
The user must also know the source hierarchy. For regulatory work, not all sources are equal. Legal text, official guidance, ECHA publications, harmonised classification entries and internal expert decisions should not be treated as interchangeable.
The user must define the boundaries. AI should be told whether it may infer, whether it must quote sources, whether it should identify uncertainty, whether it should separate facts from interpretation and whether it should produce a final answer or only a review checklist.
Most importantly, the user must be able to judge the output. AI is most useful in the hands of people who already understand the field. A regulatory expert can detect when an answer is outdated or legally imprecise. A non-expert may see only a polished paragraph.
This is why AI training should not be limited to‘ how to write prompts’. Companies should train employees to use AI in context: how to verify outputs, protect confidential data, document assumptions, select reliable sources, escalate uncertainty and decide when AI should not be used at all.
This is not only a question of efficiency, but also of governance. Companies should know which AI tools they use, with what data, for what purpose, and under what level of human review.
SDS authoring
SDSs are a good example of both the promise and the risk. At first glance, SDS authoring looks like an ideal AI use case. SDSs are structured documents that use repeated terminology and require consistency between sections.
They involve multilingual communication, standard phrases, exposure controls, first-aid instructions, transport information and regulatory references. AI can clearly help with harmonisation, wording, translation support, consistency checks and first-pass drafting.
SDS authoring is not creative writing. It is a legal and technical communication document. It must reflect the substance or mixture classification, uses, exposure conditions, regulatory context and available data. AI cannot simply guess what protection equipment is appropriate. It must be guided by someone who understands why physical form, route and duration of exposure, conditions of use, workplace scenario and legal requirements all matter.
For example, recommending gloves is not just a language task. Which material? For what breakthrough time and what duration of contact? Is the worker handling a liquid, powder, aerosol or solid article? Is the process open or closed? Is local exhaust ventilation present? Is respiratory protection required under the conditions described? These questions should not be delegated blindly to a language model.
However, this does not mean AI has no role. It can help the expert think through the structure, identify missing assumptions, compare wording, check internal consistency, prepare multilingual drafts and flag where more data is needed. The key is that AI should support the SDS author, not become the SDS author of record.
A competitive advantage
Some companies are already integrating AI into chemical compliance software, SDS authoring systems, data collection tools and regulatory monitoring workflows. This is not because AI can replace the whole job but because it can make the workflow faster, more searchable, more structured and easier to manage.
For regulatory consultants, the real competitive advantage lies in combining AI-supported efficiency with human expertise. AI can help us process information faster, organise data more clearly and identify potential risks earlier. But clients still need what AI cannot provide on its own: regulatory judgement, context, accountability and practical recommendations adapted to their specific products, supply chains and obligations.
For employees, the message should not be fear. AI will not automatically make regulatory professionals obsolete. Those who learn to use AI well may become significantly more effective than those who ignore it. A competent person can use AI to process information faster, test ideas, prepare drafts, challenge assumptions and focus more time on judgment.
For companies, investment does not mean only purchasing tools. It also means investing time, training and governance. Employees need space to experiment safely, learn what works, understand the limits and build internal rules. A company that blocks AI entirely may slow its own learning. A company that adopts AI without control may accelerate its mistakes.
Conclusion
AI will not remove the need for expertise in chemical compliance. In fact, it makes expertise more important and helps to amplify it. The future is not a choice between human experts and AI systems. The more realistic distinction is between professionals who know how to use AI critically and those who use it blindly or not at all.
AI in chemical compliance is not a threat or an advantage by default. It becomes one or the other depending on the human expertise, governance and verification behind its use. ●
Yana Trubitsyna
REGULATORY QUEEN
REGARTIS k + 420 777 352 525 J yana @ regartis. com j www. regartis. com
JUL / AUG 2026 SPECCHEMONLINE. COM
39