The Defense Advanced Research Projects Agency (DARPA) has issued an initial program solicitation titled "Kallisti," aimed at developing innovative strategies for national security decision-makers to effectively deter or incentivize actions by adversaries. The Kallisti program focuses on creating an algorithmic theory of mind to model adversaries’ situational awareness, assess their risk and reward structures, and predict future behavior in national security contexts. Proposals must be submitted by various dates leading up to a final written and oral proposal due date estimated for April 4, 2025. The program will run for 30 months, divided into three sprints, with multiple performers expected to be selected based on meeting specific goals. Proposers interested in more detailed information must request a Controlled Unclassified Information (CUI) addendum by February 14, 2025, and submit abstracts by February 28, 2025. The solicitation invites participation from businesses and research institutions capable of fulfilling these requirements. This solicitation reflects DARPA's continued commitment to advancing defense technologies and enhancing strategic decision-making capabilities.

The Kallisti Program Solicitation by the Defense Advanced Research Projects Agency (DARPA) seeks innovative proposals to enhance national security strategies through developing an algorithmic theory of mind that models adversaries’ situational awareness and predicts their future behavior. The solicitation encourages submissions from diverse entities, including small businesses and research institutions, and anticipates multiple awards. Key dates include the abstract submission deadline (February 28, 2025) and oral presentations scheduled for March 6-11, 2025. Proposals will undergo evaluations based on specified criteria listed in the Controlled Unclassified Information (CUI) addendum, which must be requested ahead of the deadline. This program is organized into three sprints over 30 months, where performers must meet specific goals at each phase to continue. The program emphasizes collaboration between algorithms and human expertise to model potential actions and outcomes in national security scenarios.

The document details a Q&A session from the DARPA Kallisti Industry Day, focusing on the development of a simulation environment for evaluating adversarial decision-making processes. Key points include DARPA's collaboration with AFRL to create a simulated environment for performance testing based on AI and modeling techniques. Initial data will be simulated due to insufficient real-world data, and DARPA encourages creativity in proposing solutions that achieve program objectives. The fidelity of the simulation will incorporate elements such as geography and military capabilities, while addressing behavioral models including irrational decision-making. Participants are advised that data poisoning is not part of the objectives, and simulators will be employed to aid in evaluating actions across various military and geopolitical scenarios. Milestones and funding management during the program emphasize iterative evaluations, with performers given the freedom to adapt their methodologies. The overarching goal is to develop an algorithmic theory of mind capable of inferring adversarial behavior in complex environments. This initiative encapsulates DARPA's commitment to integrating advanced AI methodologies for enhancing defense capabilities.

The document outlines the responses to questions submitted during the DARPA STO Kallisti Industry Day, focusing on the Kallisti program, which aims to develop an advanced simulation environment for modeling human and adversarial behavior in national security contexts. DARPA will develop the Kallisti M&A and Adjudicator Environment with contributions from AFRL and other partners. The program will use simulated data only, as real-world data is not sufficient for testing. Key aspects include modeling disproportionate human behavior, including irrationality, and the integration of cultural factors within the simulations, which will use generalized cultural representations. The program encourages diverse methodologies from proposers, emphasizing novel approaches and resilience to various data scenarios. Funding will be based on standard milestone payments. DARPA seeks multi-institutional collaborations and encourages participation from non-traditional entities in the proposal process. The document outlines various technical requirements, evaluative metrics, and provides answers to operational and procedural inquiries for potential performers involved in Kallisti's development. Overall, the Kallisti program represents an innovative exploration into AI's capability to model complex behaviors relevant to national security.