DFOP0010795 Q&A Response (1).docx

Response to Questions on NDF Funding Opportunity: DFOP0010795

1. Can we submit a proposal for part of the scope of work (and associated funding)? Specifically, will proposals that cover one of the two activities be considered?

Proposals that cover one or both activities will be considered.

2. For Activity I, is there a number of workshops/training exercises that the department is hoping for? Is there a preference for in-person or virtual events? Is there a timeline for when these events should be held?

There is no minimum number of workshops/training events that should be performed for Activity I. NDF prefers all activities to be performed in-person, but exceptions will be considered. NDF prefers all exercises performed underfunded proposals be concluded within 18 months of award.

Applicants should consider the objectives, size of the community of interest, and budget ceiling of the funding opportunity and scope their proposal, the quantity of workshops, and content of workshops accordingly.

3. What constitutes a non-US AI developer (i.e., private entity, government entity, etc.)? What does the pilot work with the proposed non-US AI developer involve? What type of AI system is being developed by the non-US developer (e.g. classifier, large language model, AI agent, etc.) and how will the AI system be used (e.g. commercial product, military/defense, critical infrastructure, government use, internal to the organization, etc.)?

Applicants should consider non-US AI developers to include entities that produce AI systems and are based outside of the United States of America, including open-source, privately owned, or governmental. After reviewing the objectives and budget ceiling of the funding opportunity, applicants should consider which types of AI systems would pose the greatest WMD-related risks if unsecured. Applicants should then identify in their proposal which AI systems, use-cases, and developers they intend to include for Activities I and/or II.

4. Is the NOFO geared more towards guardrailing of AI models (e.g., alignment of models and mitigating the risk that their outputs can be used for malicious purposes) or towards traditional cybersecurity for AI models (which may or may not be aligned) and the ecosystem that goes into building such models? Are both in scope?

This NOFO primarily seeks proposals that assist with traditional cybersecurity for AI systems and the ecosystem that goes into building such models. However, the security during AI system research and development, evaluation, deployment, or other stages may benefit from unique approaches or perspectives beyond the traditional. Proposals that incorporate nontraditional approaches as a supplement rather than a replacement to traditional cybersecurity will be considered.

5. Is the primary concern of the work in the NOFO on AI models that are openly available in the public domain and can be misused, or on proprietary models that are meant to be secured within government organizations?

The primary concern of this NOFO is mitigating security risks related to loss, theft, or diversion of AI systems and their components. This includes open-source, privately owned, or governmental AI systems. Applicants may include some or all of these entities in their proposal. Applicants should determine which of these communities to engage based on the scope and qualities of their proposal.

6. Is there a particular class of WMDs that are of primary concern (e.g., biological or chemical weapons vs. nuclear weapons)?

Competitive proposals will address cross-cutting security risks that affect all WMD classes or risks that are unique to a subset of WMD.