Overview of the BAA
The Defense Advanced Research Projects Agency (DARPA), through its Information Processing Techniques Office (IPTO), has officially released the Broad Agency Announcement (BAA) for the Decentralized Artificial Intelligence through Controlled Emergence (DICE) program (HR001126S0010). Following the Special Notice issued earlier this year, the formal solicitation now invites proposals to develop decentralized AI architectures capable of supporting scalable, adaptive, and resilient collectives of heterogeneous AI agents. The BAA provides the first detailed look at DARPA’s technical vision for DICE and outlines how organizations can participate in the program.
DICE seeks to move beyond today’s centrally orchestrated multi-agent systems (MAS) by enabling AI agents to self-organize, coordinate, and adapt through local interactions while remaining aligned with human intent. DARPA’s goal is to create AI collectives capable of executing sustained, long-horizon missions in contested environments while maintaining predictability, mission focus, and resilience to failures or adversarial compromise.
Technical Focus Areas
DICE is organized into three Technical Areas (TAs) designed to enable scalable, adaptive, and resilient collectives of AI agents that can operate in contested environments without relying on centralized control:
TA1: Decentralized Coordination and Consensus
TA1 focuses on enabling AI agents to self-organize, allocate tasks, share information, and adapt to changing conditions through peer-to-peer coordination and distributed consensus mechanisms. DARPA is interested in approaches that improve scalability, adaptability, and resilience compared to traditional orchestrated MAS.
TA2: Local Inference Control
TA2 seeks methods to maintain role coherence and mission alignment across long-duration missions. Areas of interest include activation steering, agent memory editing, context engineering, and other inference-time control approaches that help agents remain aligned while preserving flexibility and creativity.
TA3: Testing and Evaluation
TA3 will develop simulation environments, use cases, and evaluation frameworks used to assess DICE technologies and compare them against state-of-the-art (SOTA) centrally orchestrated MAS in Phase 1.
Together, these efforts span disciplines including artificial intelligence, control theory, formal methods, and game theory, with an emphasis on revolutionary advances rather than incremental improvements. Proposals pursuing the core technology effort must address both TA1 and TA2 together, while TA3 will be competing separately.
Key Dates
DARPA has established the following schedule for the DICE solicitation:
- Posting Date: June 10, 2026
- Proposal Abstract Due Date: June 30, 2026, at 2:00 PM ET
- Question Submission Deadline: August 18, 2026, at 2:00 PM ET
- Proposal Due Date: August 25, 2026, at 2:00 PM ET
DARPA strongly encourages the submission of proposal abstracts in advance of a full proposal submission.
Phases and Additional Information
DICE is structured as a 36-month effort spanning three phases. Phase 1 focuses on demonstrating the advantages of decentralized AI collectives over SOTA centrally orchestrated MAS. Phase 2 shifts to adversarial robustness, evaluating how decentralized collectives respond to compromised agents, deceptive information, and other disruptions. Phase 3 emphasizes scalability, with performers expected to support complex missions involving larger numbers of agents and interactions while maintaining adaptability and resilience.
If successful, DICE could establish a new framework for decentralized AI systems capable of coordinating thousands of specialized agents in dynamic environments while remaining resilient, adaptable, and aligned with mission objectives. More broadly, the program reflects DARPA’s interest in developing AI systems that can operate effectively in complex, contested settings where centralized control may be impractical or vulnerable.
If your company is considering applying for federal funding, your journey starts here. EverGlade is a national advisory firm helping innovators navigate the federal funding ecosystem. We support companies across the funding lifecycle, from early-stage strategy through proposal development, negotiations, and post-award execution, connecting breakthrough innovation with non-dilutive funding.
For additional information, schedule a free consultation with our team.
