Product / Case study · IBM Research
4.3× faster PFAS screening at IBM.
Every chemical product that enters an IBM Research lab is reviewed first. Safer Materials Advisor took over the slow part: reading safety data sheets and checking every ingredient against lists of restricted substances.
Results
faster review per supplier document, timed by a chemical coordinator on her own workflow
identification accuracy on hundreds of expert-labeled documents
Caught errors that manual review missed, and flagged compounds before they were regulated
The challenge
More than 4,500 chemical products, one safety data sheet at a time.
Over about two decades, IBM Research's Yorktown Heights lab has cleared more than 4,500 chemical products. Each request lands with a chemical coordinator, who reads the supplier's safety data sheet, marks it up and checks every ingredient against internal and external lists of banned and restricted substances.
PFAS make this especially hard. They come in thousands of varieties, the definitions differ between regulators, and the rules from the US EPA and others keep changing. When an ingredient is restricted, the chemical coordinator may approve the product with conditions or look for a substitute.
Chemical coordinators also report chemicals of concern to IBM's Chief Sustainability Office and to several federal agencies, and meet local officials each year about chemical use and disposal.
The solution
Four AI steps run on every safety data sheet.
The chemical coordinator points Safer Materials Advisor to the supplier documents. It does the reading and cross-checking and shows its evidence. The chemical coordinator takes the final decision.
- 01 · Read
Parse the document
Turns safety data sheets in any layout into structured, machine-readable composition data.
- 02 · Reveal
Find what's hidden
When ingredients are described vaguely or concentrations don't add up to 100%, an AI agent estimates how likely PFAS is present.
- 03 · Check
Match the rules
Recognizes PFAS by name, registry number or molecular structure, and checks which regulations apply for the location and use and whether the substance is on IBM's internal lists.
- 04 · Assess
Score the hazard
A chemistry foundation model assesses toxicity and environmental persistence, citing its sources and reporting its confidence.
Built for expert review
The system flags and explains; it doesn't certify a product as PFAS-free on its own. Every result carries its reasoning, so the chemical coordinator reviews evidence instead of hunting for it.
Keeps up with the rules
Users can add new PFAS definitions and regulations as they appear. The system reads them and adds them to its set of rules automatically.
Works with existing tools
Works with the tools the team already uses, including IBM's compliance database, so results land where they already work.
The rollout
From one chemical coordinator to production.
We built the system alongside the people who would use it, and measured the difference on real work.
- Step 1
User study
We studied how the team in IBM's Chief Sustainability Office reviews chemical products: their workflow, their sources, and the evidence they need to decide.
- Step 2
Prototype on real reviews
A chemical coordinator used a prototype in her daily work while we observed. Her timed comparison showed she was about 4.3 times as productive.
- Aug 2025
In production
Live in IBM's Chief Sustainability Office. Other team members adopted it.
- 2025
Recognized
Winner of IBM's inaugural Sustainability Prize, celebrated by ringing the closing bell at the New York Stock Exchange. Covered by Manufacturing Dive.
- Next
Moving to ChematIQs
IBM keeps using Safer Materials Advisor as the deployment moves to ChematIQs. Safer alternatives come in the next release.
Read more
- A faster way to screen supply chains for harmful chemicals ↗IBM RESEARCH BLOG · JUNE 2025
- Safer Materials Advisor project page and video ↗IBM RESEARCH
- IBM develops AI-powered PFAS screening tool ↗MANUFACTURING DIVE · JUNE 2025
- Mitigating the environmental harm of PFAS "forever chemicals" ↗IBM RESEARCH BLOG · APRIL 2024
Results are from IBM's internal deployment. IBM and IBM Research are trademarks of International Business Machines Corporation.