The UN Globally Harmonized System (GHS) defines the hazard classes (physical, health, environmental), the categories inside each class, and the label elements (pictograms, signal words, hazard statements and precautionary statements). It is not a law in itself; it is the template each region adopts. Understanding GHS well is the fastest way to understand every regional regulation on top of it, because they are mostly the same building blocks with local additions.

Regulation (EC) No 1272/2008, the Classification, Labelling and Packaging Regulation, is the EU implementation of GHS. UK CLP is the near-identical retained version after Brexit, administered by HSE with its own GB mandatory classification and labelling list. Both apply the CLP cut-off values for mixture classification, both use the same hazard classes, and both require CLP-format labels with UFI codes for products notified to Poison Centres. Divergence between EU and GB lists is where most cross-border errors happen.

OSHA's Hazard Communication Standard aligns US workplace hazard communication with GHS, with some US-specific choices (health hazards over category 4 are not adopted, some environmental hazards not covered federally). Above that sits TSCA, administered by the EPA, which governs which substances can be manufactured, imported and used in the US, plus Section 6 restrictions on specific substances. For cosmetics, MoCRA adds product listing and facility registration obligations that read from the same underlying composition data.

REACH Annex VI is the list of substances with harmonised (mandatory) classifications inside the EU. Where Annex VI applies, the harmonised entry overrides any self-classification. Every ATP (Adaptation to Technical Progress) adds or amends entries and needs to flow into your substance master. Separately, the SVHC candidate list is the roster of substances of very high concern; a substance being added does not by itself change hazard classification, but it triggers Article 33 communication obligations to downstream users at > 0.1% w/w and starts the clock on potential authorisation.

Mixture classification uses three levers: bridging principles (using data on a very similar mixture), additivity (summing contributions of classified ingredients against category cut-offs), and calculation methods per hazard class. Health hazards like acute toxicity use additivity; skin corrosion uses different cut-offs; carcinogenicity, mutagenicity and reproductive toxicity have generic concentration limits unless a specific concentration limit is set in Annex VI.

Practitioners get this wrong in two places most often: forgetting to check whether a specific concentration limit or an M-factor overrides the generic cut-off, and forgetting that an Annex VI entry on a substance can change the classification of every mixture that contains it. Both are catastrophic under audit and both are things chemical categorisation software should handle automatically.

Static classification. A mixture classified once in 2019 that has never been re-checked against ATPs to CLP and additions to the SVHC list is almost certainly wrong today.

Re-classifying only when someone remembers. Manual re-classification after regulatory updates is where most drift happens. Automated re-run against the substance master and regulatory feeds is the only defensible approach at portfolio scale.

Separate systems for classification and SDS. When the classifier and the SDS authoring tool are different products, the SDS ends up citing an older classification than the one the classifier now holds. The fix is one substance record that both read from.

No traceable rationale. Auditors ask why. If the working (cut-offs used, additivity calculation, Annex VI citation) is not stored with the record, defending it later is expensive.

Composition-driven classification for every framework you sell into, live monitoring of Annex VI, SVHC and TSCA changes against your portfolio, AI-drafted rationale with citations, one-click regeneration of SDS, label, UFI and PCN when anything changes, and full audit trail. Bolt-on classifiers do the maths; an AI operating system like Worldover holds the record, the regulation and the downstream documents as one model, so the whole downstream pipeline stays in sync.