US AI agents for critical minerals recovery: process design notes for engineers
Reviewed by Tom Sullivan
First reported on MINING.com
30 Second Briefing
US Department of Energy researchers at Pacific Northwest National Laboratory have built CICERO, a semi-autonomous system that uses AI agents, a liquid-handling robot, a sample-handling device and two analytical instruments to design and run 96 mineral recovery experiments in a single day. Tested on spent permanent magnets and oil and gas wastewater, CICERO rapidly identified flowsheets to recover magnesium from brines and neodymium, praseodymium and samarium from magnet waste. The approach uses existing industrial chemistries and separation methods, offering a route to extract critical elements from waste streams in days rather than months or years.
Technical Brief
- CICERO’s workflow links a liquid-handling robot, a sample-handling device and two analytical instruments in series.
- AI agents at PNNL design experiment chemistries, sequencing and timing directly from scientific literature, not pre-coded recipes.
- Research used two industrial waste streams: spent rare earth permanent magnets and oil and gas extraction wastewater.
- For these feeds, agents autonomously selected target elements based on composition and value.
- Economic and industrial feasibility is evaluated alongside separation performance, screening out flowsheets unlikely to scale.
- Scope is currently bench-scale and limited to chemistries already documented in the literature and industrial practice.
- For operations, CICERO offers a rapid route to tailor flowsheets to site-specific waste brines and magnet scrap.
Our Take
The US Department of Energy’s use of CICERO for critical minerals like neodymium and praseodymium aligns with its recent TRACE-Ga and USA Rare Earth initiatives, signalling a coordinated push to de-risk multiple bottleneck elements across the supply chain rather than backing isolated projects.
Cutting method-development timelines from months to days for magnesium, copper and rare earths suggests mid-tier producers and recyclers in the USA could move pilot flowsheets to demonstration scale much faster, which in turn may compress the window for proprietary process advantages held by traditional labs.
Within our mining database, only a handful of critical minerals pieces explicitly reference AI or artificial intelligence, so the DOE–PNNL CICERO work stands out as one of the first attempts to operationalise AI agents directly into lab-scale metallurgical test design rather than just mine planning or exploration targeting.
Prepared by collating external sources, AI-assisted tools, and Geomechanics.io’s proprietary mining database, then reviewed for technical accuracy & edited by our geotechnical team.
Related Articles
Related Industries & Products
Mining
Geotechnical software solutions for mining operations including CMRR analysis, hydrogeological testing, and data management.
CMRR-io
Streamline coal mine roof stability assessments with our cloud-based CMRR software featuring automated calculations, multi-scenario analysis, and collaborative workflows.
HYDROGEO-io
Comprehensive hydrogeological testing platform for managing, analysing, and reporting on packer tests, lugeon values, and hydraulic conductivity assessments.
GEODB-io
Centralised geotechnical data management solution for storing, accessing, and analysing all your site investigation and material testing data.