Units
1. Introduction to Artificial Intelligence in Chemistry: Roles, Boundaries, Validation and Ethics 2. Molecular Representation and Chemical Structure: Validation with SMILES, InChI and RDKit 3. Reaction Prediction and Mechanism: Product, Condition and Adverse Reaction Prediction 4. Spectrum Interpretation Support: NMR, IR and Mass Spectrometry 5. Literature Review and Synthesis Planning: Source, Procedure and Roadmap 6. Chemical Data Analysis and Python: pandas, Stoichiometry and Calibration 7. Molecular Property Prediction and QSAR: Resolution, pKa and Model Limits 8. Laboratory Documentation: Experiment Notebook, ELN and Reproducibility 9. Safety and Hazard Assessment: GHS, SDS and the Limits of Artificial Intelligence 10. Confirmation, Hallucination, and Scientific Integrity 11. Human in Experimentation: Ethics, Privacy, Responsibility and End-to-End Workflow
Unit 9 / 11

Safety and Hazard Assessment: GHS, SDS and the Limits of Artificial Intelligence

Gains:

  • Understanding that the source of security truth is SDS and GHS and that artificial intelligence cannot have the last word in security
  • Ability to screen for incompatible reagent combinations in every procedure and cross-validate with SDSs
  • Ability to determine PPE at material level and have the procedure approved by an experienced person/safety officer

Safety and Hazard Assessment: Limits of GHS, SDS and AI

In chemistry, no issue is more important than safety. A molecular weight error would disrupt the experiment; A security error can cost human life. AI can speed up safety documentation: produce hazard summaries, recommend personal protective equipment (PPE), draft waste disposal note, issue risk assessment form. But there is one absolute rule: AI can never have the final say on security. The safety decision rests with official sources (SDS — Safety Data Sheet, GHS — Globally Harmonized System of Classification of Chemicals) and a competent person. In this unit we will learn to use AI as an aid in security but never to rely on it.

Basic concepts

  • SDS (Safety Data Sheet): Official 16-section document provided by the manufacturer of each chemical. Includes hazards, first aid, storage, disposal. Security is the source of truth.
  • GHS: International system that divides chemicals into hazard classes (flammable, corrosive, toxic, oxidizing, etc.) and displays them with pictograms.
  • H and P statements: H (Hazard) hazard statements (e.g. H314: severe skin burns), P (Precaution) precaution statements (e.g. P280: wear protective gloves).
  • PPE: Personal protective equipment — goggles, gloves, apron, respiratory protection if necessary.

AI is good at explaining these concepts and summarizing an SDS text. But it is dangerous to rely on hazard information provided by AI "from memory" without obtaining the actual SDS of a chemical from the manufacturer; It may be incomplete or incorrect.

Caution: It is never enough for the AI ​​to say “safe” or “not hazardous” for a chemical. Obtain the official SDS for each chemical from the manufacturer/supplier and read hazard, incompatibility, disposal information there. AI is not a replacement for SDS.

The most critical danger: incompatible combinations

While some chemicals are relatively safe alone, they are dangerous together. Classic examples: oxidizers + organic substances (fire/explosion), acids + hypochlorite (chlorine gas), water + reactive metals (hydrogen/heat), acids + cyanide salts (toxic gas). AI knows most of these combinations, but it does not guarantee all of them, and can sometimes innocently present a dangerous suggestion. Each procedure should be answered by asking “are there any incompatible reagents here?” It is your job to scan with your eyes.

Step by step: AI-powered safe preparation

  1. Create a list of reagents: List each chemical used in the experiment.
  2. Get actual SDSs: Download SDS from manufacturer/supplier for each.
  3. Summarize to AI: paste SDS text and extract hazard/PPE/disposal summary.
  4. Incompatibility screening: Ask the AI ​​"are there any dangerous combinations between these reagents?" ask; verify the result with SDSs.
  5. Risk assessment form: Let AI produce a draft, you fill it with real information and have it approved by the official.
  6. Human approval: Do not perform the procedure without an experienced person/safety officer approving it.

Four copyable templates

1) SDS summary (you provide the text):

Below is the SDS text for a chemical:[SDS TEXT]Task: Omit: GHS classes, H and P statements, required PPE, storage condition, incompatible substances, spill and first aid summary, waste disposal method. Add information NOT in the text; otherwise write "not specified in SDS".

2) Incompatibility scan:

The following chemicals are used together/sequentially in the experiment:[LIST + short procedure]Task: Are there any known hazardous combinations (oxidizer+organic, acid+hypochlorite, water+reactive metal, etc.) among these substances?Describe each possible risk.WARNING: This is a PRELIMINARY screening; I will verify the SDS of each item.

3) Draft risk assessment form:

Task: Produce a laboratory risk assessment form template. Fields: job description, substances and hazard classes used, routes of exposure, required PPE, engineering measures (fume hood), emergency procedure, waste disposal, approving person/date. Leave each field blank to be filled in.

4) PPE and precaution recommendation:

Procedure: working with concentrated nitric acid, in fume hood.Task: List appropriate PPE (including glove type), engineering precautions and first steps in the event of a spill.Note: I will confirm glove material selection with the SDS compatibility chart.

Weak prompt / Strong prompt

Weak:

Is it safe to work with this chemical?

The AI ​​may respond from memory, such as “generally safe”; this is incomplete and misleading, omitting the actual SDS.

Strong:

Below I paste the SDS text for chemical Adding a security claim that is not in the text. Result: I will have this summary approved by the security officer.

Difference: real SDS was taken as basis, AI just summarized, fabrication was banned, human confirmation was brought into the flow.

Hazard classes and typical precautions

GHS class

example

Main risk

Typical precaution

flammable liquid

acetone, ethanol

fire

Away from flame, sparkless

corrosive

Concentrated acid/base

Skin/eye burn

Face shield, suitable gloves

toxic

Cyanide salts

poisoning

Fume hood, closed system

oxidizer

Nitric acid, perchlorate

fire/explosion

Store separately from organics

sensitizer

some amines

allergic reaction

Avoid skin contact

mini cases

Case 1 — AI's secret dangerous suggestion. A user requested a surface cleaning procedure; YZ indirectly suggested the sequential use of an acidic solution with a product containing hypochlorite—risk of chlorine gas. The user kept looking at the "incompatible substances" section in the SDSs. Lesson: Cross the AI ​​recommendation with the SDS mismatch section.

Case 2 — Missing PPE. YZ said "gloves and goggles" for working with a concentrated acid but did not specify the type of gloves; Some glove materials do not withstand that acid. The user selected the correct glove material from the SDS's compatibility chart. Lesson: Verify PPE recommendation from SDS at material level.

Case 3 — SDS summary added value. One student had AI summarize a 12-page SDS; hazard, PPE and disposal were reduced to 1 page and read quickly before the experiment. The summary was cross-checked with the original SDS, it was correct. Lesson: AI SDS is safe and valuable in summarizing because you provide the text.

Common mistakes

  • Just asking the AI for security. Making a "safe/not safe" decision without actual SDS.
  • Not scanning for incompatible combinations. Ignoring that individually safe substances can be dangerous together.
  • Not specifying PPE at the material level. “Gloves” are not enough; Which material, what thickness?
  • Not taking into account scale. What is insignificant heat on a mg scale can be dangerous on a mole scale.
  • Skipping waste disposal. Although the reaction is safe, proper disposal of the waste is a separate hazard.
  • Bypassing human approval. Performing the procedure without approval from an experienced person.
Tip: Apply the “zero trust” principle in security: Treat every AI security claim as unconfirmed until proven otherwise. The source is SDS, the decision maker is human.

In summary

  • Safety is the highest priority in chemistry; Here AI can never have the last word.
  • The source of actual safety information is SDS and GHS; When AI gives from memory, it may be incomplete/incorrect.
  • The most critical risk is incompatible combinations; scan each procedure with this eye.
  • Use AI in SDS summarization, form drafting and preliminary non-compliance screening; verify all with SDS.
  • Specify PPE at material level, take into account scale and waste disposal, obtain human approval.

Application task

Select three chemicals you are working with. Obtain the actual SDS for each from the supplier. Have the AI ​​summarize the SDS texts (hazard, PPE, incompatible substances, disposal). Then have the AI ​​scan for incompatibilities using all three together and cross the result with the "incompatible substances" sections of the SDSs. Fill out a risk assessment form. Try to find at least one point of non-compliance or missing PPE; Write a plan to have an official approve the form.

checklist

  • [ ] I get the real SDS for each chemical, I don't rely on AI.
  • [ ] I cross-check the AI's security summary with SDS.
  • [ ] I scan the procedure for incompatible combinations.
  • [ ] I verify PPE at material level from SDS.
  • [ ] I consider scale and waste disposal.
  • [ ] I have the procedure approved by an experienced person/security officer.
Previous Next