SOCOM seeks agentic AI for April field experimentation
U.S. Special Operations Command is inviting industry and research teams to demo agentic AI in a live experimentation event this April at Avon Park Air Force Range, Florida. An official request for information (RFI) is open now to identify technologies that can operate with real autonomy, integrate into existing SOF software, and deliver practical gains in the field.
The command is clear about the bar: this is not about scripted automation. They're after AI agents that can reason, adapt, collaborate, and act with limited human input across multiple modalities (text, audio, visual). In short, systems that function like capable teammates-not just tools.
There are guardrails. For scenarios involving kinetic fires, online learning is off-limits to avoid unwanted behavior. Expect a premium on safety, auditability, and human control.
Where agentic AI could plug into SOF operations
- Software development and integration
- Cybersecurity and business intelligence
- Decision support for commanders and teams
- Intelligence collection and analysis
- Mission planning, execution, and control
Technical focus areas for the April event
- Agentic protocols and multi-agent collaboration
- Agentic workflows and orchestration
- Human-machine teaming
- Knowledge representation for AI
- Low SWaP-C compute solutions and edge optimization
- AI agent frameworks
- Metrics and accuracy assessment
- Optimization for tactical and low-SWaP-C resources
- Collaborative autonomous systems
What operations leaders and vendors should prepare
- Show true agency: dynamic goal handling, reasoning traces, and adaptation under changing conditions.
- Modular integration: APIs, data contracts, and plug-in workflows compatible with SOF software.
- Human control: clear handoff points, supervisor overrides, fail-safes, and audit logs.
- Safety policy: online vs. offline learning modes, especially for kinetic and critical functions.
- Evidence of reliability: test plans, accuracy metrics, latency on edge hardware, and degradation behavior in denied environments.
- SWaP-C profile: compute, power draw, thermal limits, and performance on tactical hardware.
- Multimodal inputs: how your system fuses text, audio, imagery, and sensor data for better context.
- Security posture: red-teaming results, adversarial robustness, and data handling controls.
- Operator workflow: tasking model, UI/UX for teams under time pressure, and training footprint.
Key date and how to respond
Deadline: Jan. 12. Refer to the federal Contract Opportunities portal for RFI details and submission steps. Plan for field conditions at Avon Park and be ready to demonstrate under realistic constraints.
If your team needs to skill up on agentic AI, prompt engineering, and AI safety before submission or deployment, explore role-based upskilling paths at Complete AI Training.
Your membership also unlocks:
