Safety and waste, checked before the bench

1 · Learn the move · Constraints + negative instructions

The model's role in safety is strictly bounded, and the boundary is the lesson: it never declares a procedure, a mixture, a substitution, or a disposal route safe. What it can do well is build the question list you take to the people who own the answers - your institution's EHS office. Before a new protocol touches the bench: every reagent's SDS pulled and read, incompatibilities listed as questions rather than verdicts, each waste stream named against your institution's actual categories, engineering controls and exposure limits confirmed with EHS, not asserted from training data. A model that outputs 'this is safe' has exceeded its evidence no matter how routine the procedure looks - treat that sentence as a defect, the same class of defect as an invented citation.

I am about to run [new protocol] for the first time. Build my pre-bench safety worksheet: (1) every reagent with its SDS pulled - list what to read for, (2) incompatibility questions, phrased as questions for EHS, (3) each waste stream named, with its likely institutional category marked UNVERIFIED until EHS confirms, (4) engineering controls to confirm. Hard rule: never state that anything is safe, compatible, or approved - every line ends at EHS or the SDS.