OtherVerified
Reviewed and published by trentmaziarz, July 7, 2026. Discovered and drafted by our automated research pipeline.
MITRE deployed the Federal AI Sandbox, a supercomputer powered by an NVIDIA DGX SuperPOD, to train and evaluate large-scale AI foundation models for U.S. federal agency missions. The Sandbox, located in Ashburn, Virginia, became operational in late 2024 and is accessible to federal agencies through existing FFRDC contracts.
Details
The Federal AI Sandbox provides one exaFLOP of 8-bit AI compute performance using 248 NVIDIA H100 GPUs and 9 PB of storage, representing a roughly 300-fold increase over MITRE's prior AI compute capabilities. It supports training of large language models, multimodal perception systems, and reinforcement learning decision aids. Confirmed use cases include weather forecasting in collaboration with NOAA and the National Weather Service, cybersecurity foundation model development to assist security operations center analysts, and public benefits administration AI to process complex regulations and government workflows. The Sandbox also supports satellite imagery analysis and aviation applications.
Products affected
MITRE Federal AI Sandbox
Sources & Evidence
Company Disclosure
Cite this record
Trace Foundation. (2026). MITRE: MITRE deployed the Federal AI Sandbox, a supercomputer powered by an NVIDIA DGX SuperPOD, to train and evaluate large-scale AI foundation models for U.S. federal agency missions. The Sandbox, located in Ashburn, Virginia, became operational in late 2024 and is accessible to federal agencies through existing FFRDC contracts (data as of 2026-07-07) [Data set record]. AI Trace. https://www.aitrace.org/r/practice/336f1f8f-4bd6-460b-bcbc-cdfe22bd69be. Accessed October 5, 2026.
- Stable link
- https://www.aitrace.org/r/practice/336f1f8f-4bd6-460b-bcbc-cdfe22bd69be
- Data as of
- July 7, 2026
- Last verified
- Not recorded
Other practices by MITRE
OtherMITRE deployed its AI Assurance and Discovery Lab in March 2024 at its McLean, Virginia headquarters to evaluate risks in AI-enabled systems for federal agencies and private companies. The lab uses simulated environments, AI red teaming, and large language model evaluation to identify security, bias, and performance risks before agencies adopt AI systems.OtherMITRE announced in December 2025 that it will operate two new AI Economic Security Centers on behalf of NIST, funded by a $20 million NIST investment, focused on applying AI to U.S. manufacturing productivity and securing critical infrastructure from cyberthreats. Both centers will leverage MITRE's Federal AI Sandbox and existing tools.Data AnalysisMITRE's Center for Threat-Informed Defense offers TRAM (Threat Report ATT&CK Mapper), an open-source tool that uses machine learning and large language models to automatically identify and classify adversary tactics and techniques in cybersecurity threat intelligence reports. The tool assists security analysts who previously had to map reports manually.
Have evidence about MITRE's AI practices? Submit a report.
Report a Sighting →