Request for Information (RFI) - Artificial Intelligence for Image Adjudication

Questions & Answers / Clarifications

This document provides responses to vendor questions regarding the Request for Information (RFI) for Artificial Intelligence (AI) for Image Adjudication. It clarifies aspects of the RFI, including data formats, deployment strategies, and potential procurement timelines.

• Incumbent Services: CBP previously leveraged DHS's Commercial Solutions Opening Process (CSOP) in 2023 for similar services.
• Data Format & Volume: Current LS-NII image data uses UFF2 or UFF3 standards, with approximately 50,000 vehicles scanned daily.
• Deployment Preference: Algorithms will be deployed in the CBP cloud using an internally developed execution environment, with potential consideration for edge or hybrid models in the future.
• Integration: The AI component must integrate with CBP's cloud computing environment.
• Procurement Timeline: This RFI is for investigation, not scheduled to move forward with an RFP award, but to assess AI advancements.
• Data Sharing: CBP will share UFF image files and metadata with a vendor if they decide to move forward with an RFP process. No publicly available WCO reference sets or simulators exist.
• Training Data: CBP has labeled training data available for model development and evaluation. They are exploring pathways for vendors without existing CBP Background Investigation footprints to access government data.
• Schema: CBP supports multiple models running in parallel/cascading operation and welcomes vendor proposals for draft schemas.
• Pilot Pathway: CBP anticipates a shadow-mode evaluation using real NII data, followed by testing at a single location before large-scale deployment.
• Latency: Real-time is defined as less than five seconds for AI inference.
• Data Availability: CBP curates a weekly dataset of several hundred annotated images.

This Q&A document clarifies the scope and technical requirements of the RFI, providing potential bidders with a clearer understanding of CBP's needs and expectations for AI in image adjudication. It indicates that while this RFI is for investigation, CBP may consider future procurement actions based on the findings.

Request for Information (RFI)

This RFI is issued by U.S. Department of Homeland Security, Customs and Border Protection (CBP) for information and planning purposes only. It is not a Request for Proposal (RFP) and does not commit the government to a contract. CBP is seeking information on how to best develop Artificial Intelligence (AI) and Machine Learning (ML) models to assist in Non-Intrusive Inspection (NII) image adjudication. The goal is to improve efficiency and effectiveness in NII operations, particularly for large-scale scanning of privately owned vehicles (POVs) and cargo vehicles.

CBP is interested in modular algorithms and models that can:

• Operate on large-scale NII imagery and associated metadata (emphasis on WCO UFF).
• Be adapted to CBP operational needs and constraints.
• Integrate as drop-in components within existing CBP architectures, rather than requiring new proprietary platforms.

Vendors are asked to describe:

• Developed or developing algorithm products for anomaly detection, manifest/commodity verification, conveyance modification detection, and contraband detection, including performance metrics.

• Development approaches, including model types, generalization capabilities, and limitations.

• Capabilities in image segmentation and supported image formats (including UFF).

• Technical dependencies, system requirements, and integration approaches.

• Field testing experience, operational settings, and performance results (detection and false alarm rates).

• Methods and requirements for training, verification, and deployment, including handling of incomplete or inconsistent metadata.

• How AI/ML models evolve through continuous training and performance monitoring.

• Specific contraband or conditions their AI/ML addresses.

• Supported LS-NII modalities (POVs, cargo vehicles, sea containers, rail cargo).

• Strategies for minimizing technical risk and life cycle costs.

• How industry can help standardize technology and leverage solutions for interoperability.

• Subcontracting/teaming arrangements, including small/disadvantaged businesses.

• Collaborative activities to encourage open architecture, non-proprietary AI/ML solutions.

• Questions must be submitted by 12:00 PM EST, May 15, 2026.

• Responses must be submitted via email no later than 12:00 PM EST, May 30, 2026.

• Electronic submissions are required, with files larger than 10 MB broken down into multiple files not exceeding 10 MB each.

• Submissions are limited to fifteen (15) pages, excluding certain introductory pages.

• All costs associated with responding are at the respondent's expense.

• Not responding does not preclude future participation in solicitations.

• Information provided is subject to change and is not binding on the Government.

Request for Information (RFI)

This RFI is issued by U.S. Department of Homeland Security, Customs and Border Protection (CBP) for information and planning purposes only. It is not a Request for Proposal (RFP) and does not commit the government to a contract. CBP is seeking information on how to best develop and implement Artificial Intelligence (AI) and Machine Learning (ML) models to assist in Non-Intrusive Inspection (NII) image adjudication. The goal is to improve efficiency and effectiveness in processing high volumes of X-ray images and associated metadata from large-scale (LS) imaging systems used for privately owned vehicles (POVs) and cargo vehicles.

CBP is interested in modular AI/ML algorithms and models that can:

• Operate on large-scale NII imagery and metadata (emphasis on WCO UFF).
• Be adapted to CBP operational needs and constraints.
• Integrate as drop-in components within existing CBP architectures, without requiring new proprietary data platforms or pipelines.

Vendors are asked to describe:

• Developed or developing algorithm products for applications like anomaly detection, manifest/commodity verification, conveyance modification detection, and contraband detection, including performance metrics.

• Development approach, including models/architectures used, limitations, and how they generalize.

• Capabilities in image segmentation and supported image formats (including UFF).

• Technical dependencies, system requirements, and potential incompatibilities.

• Typical architecture, training, deployment, and scaling strategies.

• Field testing experience, operational settings, and performance results.

• Methods and requirements for training, verification, and deployment, including data requirements and handling of incomplete/inconsistent metadata.

• Algorithm system integration approach, including APIs and integration with CBP viewers.

• How AI/ML models evolve through continuous training and performance monitoring.

• Specific applications of AI/ML for contraband/condition detection and supported LS-NII modalities.

• Investment, acquisition, design, and management strategies to minimize risk and costs.

• How industry can help standardize technology and maximize interoperability.

• Subcontracting/teaming arrangements, including small/disadvantaged businesses.

• Collaborative activities to encourage open architecture, non-proprietary AI/ML solutions.

• Questions must be submitted by 12:00 PM EST, May 15, 2026.

• Responses must be sent via email no later than 12:00 PM EST, May 30, 2026.

• Electronic submissions only, with files larger than 10 MB broken down into multiple files not exceeding 10 MB each.

• Page limit of fifteen (15) pages per submission (excluding specific front/back matter).

• This RFI is for Government Market Research and does not constitute a commitment to procure.

• All costs associated with responding are at the respondent's expense.

• Submissions become government property and will not be returned.