GRC joins DOE's Genesis Mission with Argonne and PNNL
GRC is a partner in two of the first projects of the Department of Energy's Genesis Mission. SPOTTER-AI, led by Argonne National Laboratory, protects AI-driven science workflows from tampered data. Portus, led by Pacific Northwest National Laboratory (PNNL), helps coding agents run scientific workflows well on HPC and cloud systems. GRC is a co-principal investigator on both and brings CLIO, the agent layer of IOWarp.
The Genesis Mission
Through the Genesis Mission, DOE is building the American Science Cloud, a national platform where researchers share datasets, models and workflows and use AI to speed up discovery. On July 22, DOE posted the list of the first projects selected for award negotiations under the mission. Both projects with GRC are on it, and work on both began on August 1.
SPOTTER-AI: guarding AI workflows from tampered data
Openness is what makes a shared science platform useful, and it is also a risk. A few tampered images or one corrupted data source can quietly skew an AI model and spread through many downstream studies, with no record of where the error entered. Conventional cybersecurity protects the infrastructure, so these attacks on the data itself slip through.
SPOTTER-AI, short for Scientific Provenance-Oriented Threat Tracing and Attribution for Genesis Workflows, works in three steps:
- Track. It records each workflow's full history, every dataset, model step and tool call, as provenance.
- Detect. It asks whether a piece of data should have mattered as much as it did to a result. A transformer model combines provenance with signals such as attention scores to tell legitimate high-impact inputs from tampered ones.
- Trace. When something looks wrong, an AI agent reasons backward through the workflow's history to find the source of the attack and describe how it happened.
Because it inspects only the suspicious parts of a workflow, SPOTTER-AI's advantage grows as workflows get larger. The team will first test it on a digital twin from one of the OPAL research projects, which predicts plant traits from images, sensor readings and genetics. Each piece is being built as a reusable American Science Cloud service.
Bogdan Nicolae of Argonne leads the project. The partners include Oak Ridge National Laboratory, Hewlett Packard Enterprise and Illinois Tech. At GRC, Anthony Kougkas and Xian-He Sun lead the work with Jaime Cernuda, contributing CLIO Core, CLIO Kit, CLIO Agent and CLIO Coder.
Portus: performance portability for agentic workflows
Agentic workflows change their execution paths as they run, generate new pieces on the fly, and must cope with computing resources that contend, fail and change. Without better orchestration, scientists would have to tune every port to a new system by hand or accept large performance losses.
Portus starts from the observation that agentic workflows are built from a small vocabulary of recurring patterns. It breaks workflows into producer-consumer building blocks and learns models of how they perform that carry over from one workflow to another. A coding agent can then use those models to find bottlenecks in newly generated workflow code, rank alternatives and map it to the resources available. The models keep learning from live telemetry, because a fixed training set cannot keep up with changing workflows and resources.
Nathan Tallent of PNNL leads the project. GRC's part covers performance models for producer-consumer data flow and a storage model for data placement, while CLIO provides context and data services to the agents. Kougkas and Sun lead the GRC work with Xiaoyang Lu and Luke Logan.
Two partnerships, one software stack
Both projects build on long collaborations. With Nicolae, GRC works on I/O for scientific AI, including PKAS at HPDC 2026. With Tallent's group at PNNL, GRC worked on DaYu and DataFlowDrs, tools that analyze how data flows through workflows. DataFlowDrs was accepted by IEEE TPDS this summer. In both projects, CLIO gives AI agents access to scientific data, tools and context on HPC systems.
Update (November 2026): Nicolae and Cernuda demonstrate SPOTTER-AI at the DOE booth (4829, station 2) at SC26 in Chicago on Wednesday, November 18, from 12:00 to 1:00 p.m. See GRC at SC26.