Attachment_1_RFI_USSOCOM_TE_Event_26-2.docx

UNCLASSIFIED

UNCLASSIFIED

Solicitation Number: USSOCOM RFI TE 26-2 Agentic Artificial Intelligence (AI)-based Modular Capabilities for SOF Notice Type: Special Notice TYPE: A–Research and Development NAICS: 541715 Research and Development in the Physical, Engineering, and Life Sciences (except Nanotechnology and Biotechnology) Publish Date: 12 December 2025 Response Date: 12 January 2026, 12:00 Noon Eastern Standard Time (EST) Synopsis:

A. INTRODUCTION: Technical Experimentation (TE) This Request for Information (RFI) is NOT a solicitation for proposals, proposal abstracts, or quotations. The purpose of this RFI is to solicit technology experimentation candidates from Research and Development (R&D) organizations, private industry, and academia for inclusion in future experimentation events coordinated by the U. S.

Special Operations Command (USSOCOM). USSOCOM invites industry, academia, individuals, and Government labs to submit technology experimentation nominations addressing innovative technologies identified in paragraph B. 4 below. The intent is to provide participants with the opportunity to gain Special Operations Forces (SOF) insight/perspective on participant technologies.

TE explores emerging technologies, technical applications, and their potential to provide solutions for future SOF capabilities. This RFI is for TE 26-2. This event occurs over the period of one week as shown below.

Dates/Theme/Location:

13 April - 17 April 2026: Agentic AI at Avon Park Air Force Range, FL.

B. OBJECTIVE:

1. TE events provide an opportunity for technology developers to interact with operational personnel to determine how their technology development efforts and ideas may support or enhance SOF capability needs. The environment facilitates a collaborative relationship between Government, academia, labs, and industry to promote the identification and assessment of emerging technologies.

2. The deadline for nomination package(s) is identified in the “Response Date” above. After reviewing the TE nomination submissions, the Government may invite select candidates to demonstrate their technologies at the USSOCOM sponsored TE event. Experiments are conducted as defined in section A and explore emerging technology solutions and revolutionary improvements in relevant technologies. Materiel solutions brought to the event should be between a Technology Readiness Level (TRL) of 3 to 6. Proposed experiments may be conducted between a half day and two days in duration and may be conducted in unimproved expeditionary-like conditions. There is no intention on the part of USSOCOM to purchase or procure equipment based solely on participation in the TE.

3. Experimentation Focus: The primary intent of this event is to highlight technologies that support the modular integration of Agentic AI-based technologies into existing SOF capabilities.

Problem Statement: SOF requires Agentic Artificial Intelligence (AI)-based technologies that can collaborate to perform tactical or operational tasks, create and support workflows that can be modularly integrated into existing SOF software architectures.

Definition of Agentic AI: Shavit et al.[endnoteRef:1] state, “Agentic AI systems are characterized by the ability to take actions [that] consistently contribute toward achieving goals over an extended period of time, without their behavior having been specified in advance” [emphasis added]. Agentic AI refers to AI systems capable of autonomous decision-making, reasoning, and adaptation. Agentic AI-based capability refers to a class of AI agents that can perceive and interact with multiple modalities, such as visual, auditory, and textual inputs and automatically respond to conditions, autonomously make decisions, and perform complex tasks independently, without human intervention or with limited human intervention. Agentic AI-based systems utilize a multimodal framework to interpret a broader context, enhancing interactions by combining various data types. For instance, multimodal AI agents can parse language, recognize tone, and interpret visual data simultaneously, leading to more accurate and context-aware responses. This approach is crucial for improving human-machine interaction and adapting to diverse environments to facilitate a more fluid and responsive engagement with digital systems. [1: Shavit, Y., S. Agarwal, M. Brundage, S. Adler, C. O’Keefe, R. Campbell, T. Lee, P. Mishkin, T. Eloundou, A. Hickey, K. Slama, L. Ahmad, P. McMillan, A. Beutel, A. Passos, and D. G. Robinson. “Practices for Governing Agentic AI Systems.” OpenAI, https://cdn.openai.com/papers/practices-for-governing-agentic-ai-systems.pdf, accessed April 2025]

Agentic workflows can replace some or enhance other mission workflows, particularly when combined with Human-Machine Teaming (HMT). Automation does not qualify as an agentic system. Automation refers to systems or processes that operate automatically based on predefined rules, instructions, or algorithms. These systems do not require human intervention once they are programmed. They operate in a predictable, rule-based manner.

USSOCOM is not looking for automation in this TE. Agentic solutions exhibit autonomy and decision-making capabilities. These systems can adapt, learn, and make choices based on their environment or goals, often mimicking human-like agency. They operate with a degree of independence and adaptability. However, in some cases (e.g., kinetic fires) online learning is not allowed since it may lead to undesired behavior.

In the context of the U.S. government, incorporating agentic AI-based capability is expected to enhance the decision-making pipeline to deliver effects across various sectors by enabling systems to reason, remember, and access external tools and resources effectively. Applying technology to SOF architectures is intended to streamline operations, improve oversight, and potentially address challenges in areas like intelligence and cybersecurity. The field is closely linked to agentic automation, also known as agent-based process management systems, when applied to process automation. Potential SOF Agentic AI applications include software development and integration, cybersecurity and business intelligence, decision support, intelligence gathering and analysis, mission planning, mission execution, and mission control. An Agentic AI agents may control other agents.

Participants need to present their demonstrations or experiments in one of the various mission contexts relevant to SOF. Using generic scenarios (especially ones out of the military context) often make it difficult for SOF personnel to make the logical leap to scenarios more specific to their missions.

Reference: CSIAC (Cyber Security and Information Systems Information Analysis Center) TECHNICAL INQUIRY (TI), RESPONSE REPORT, Agentic Artificial Intelligence: Strategic Adoption in the U.S. Department of Defense, Report Number: CSIAC-BCO-2025-701, Completed June 2025[endnoteRef:2] [2: https://csiac.dtic.mil/wp-content/uploads/2025/09/TI-Response-Report_CSIAC_Agentic-AI-Strategic-Adoption-in-the-DoD_.pdf]

Note: The following items are not intended to be comprehensive treatment of the problem areas. The intent is to focus on specific technologies, tools, and solutions that USSOCOM can implement in a cost-efficient manner with low operations and sustainment costs. This is NOT limited to Unmanned Systems (UxS), since there are many use cases for agentic systems besides UxS. In fact, non-UxS use cases are of particular interest, including decision support, mission planning and rehearsal, intelligence, data analysis and workflows, etc.

4. Technology areas to explore during the event include the following:

This paragraph lists some Agentic AI USSOCOM is interested in exploring. There are other possible explorations, so USSOCOM is open to respondent input with respect to Agentic AI topics not listed here.

4.1. Agentic Protocols (e.g., Model Context Protocol (MCP), Agent-to-Agent (A2A) communication, and Agent Network Communication (ANC), etc.). Performers’ solutions should leverage emerging agentic protocols. Performers should show an architecture diagram and provide details about the Tools, Prompts, or Resources exposed on an MCP Server, for example.

4.2. Agentic Workflows or Orchestration: These are systems where AI agents are given autonomy to manage and execute tasks, often involving multiple steps and potentially complex reasoning, with minimal human oversight. These workflows go beyond traditional automation.

4.3. Human-Machine Teaming (HMT)

4.3.1. Various Human-Machine Interfaces (HMI) such as graphical, tactile, auditory, voice to text, text to voice, etc.

4.3.2. Areas include enhanced decision support (weighted Course of Action (CoA) redirection or realignment of resources for ongoing actions), adaptive task allocation, natural language interaction, autonomous systems integration, training and simulation, cognitive augmentation, risks to platform/risks to mission determinations, etc.

4.3.3. Note: These apply beyond unmanned systems, and USSOCOM is interested in UxS and non-UxS solutions.

4.4. Knowledge Representation for AI: To effectively facilitate agentic AI in military operations, Knowledge Representation (KR) must be designed to support the AI's ability to understand, reason, and act autonomously. This requires augmenting or enhancing common data model (schema)-based messages to ensure they can represent complex, multi-domain knowledge in ways that are machine-readable, interoperable, and actionable (e.g., semantic web technologies, preferred ontologies, vector stores, knowledge and semantic graphs, uncertainty measures, etc.). Ontological relationships enable AI inference as the bullets below illustrate:

· The AI identifies that a RADAR system is a material entity with the role of providing surveillance.

· It recognizes that the RADAR system is currently in a state of being offline due to the process of maintenance.

· The AI connects the RADAR system’s offline state to the base’s disposition of operational readiness, which is now compromised.

· The AI observes the process of increased enemy aircraft activity in the region (from other sensors) and infers that the lack of RADAR surveillance increases the risk of undetected threats. The agent recommends: (1) Deploying alternative surveillance systems (e.g., drones or satellite imagery) to compensate for the offline RADAR and (2) Implementing heightened alert levels and defensive measures.

For example, if the Automatic Target Recognition (ATR) reports a “tank,” it should include semantics and ontological metadata to differentiate this object as a vessel for liquids or a tracked vehicle. Ontological “is a weapons system” supports further processes. The ontological relationships enable the AI to connect seemingly unrelated pieces of information (e.g., vehicle, location, enemy faction, sabotage) and infer a potential threat. Without the ontology, the AI would treat the data as isolated facts and fail to generate actionable insights.

· Using the example above, an agentic AI solution would use above information combined with other intelligence like location and movement/direction to provide leadership with a weighted CoA along with automatic notifications to vulnerable forces of potential impending threat.

4.4.1. Unstructured Document Analysis/Integration

4.4.1.1. Open Source (e.g., Docling) and COTS document parsers and datastores (e.g., vector database) to create knowledge representations amenable for AI. These technologies prepare data for AI (e.g., DocTags formatting, Optimized Table-Structure Language (OTSL)). These technologies facilitate agentic systems.

4.5. Low-SWaP-C Solutions low Size, Weight, and Power-Compute (SWaP-C) systems can run on disadvantaged edge compute that span from onboard UxS devices, laptops, and ruggedized server racks that present data stores that include AI-friendly knowledge representations.

4.6. AI Agent Frameworks and Workflow Automation Platforms such as n8n, CrewAI, LangChain, LangGraph, Rivet, & LlamaIndex that facilitate building agentic orchestration or workflows.

4.7. Metrics and AI Accuracy Assessment and tools and methodologies to implement them (i.e., What should USSOCOM implement to assess prospective and deployed AI algorithms and models?) – F1-score, confusion matrix, Area Under the Receiver Operating Characteristic (ROC) Curve (AUC-ROC), Bilingual Evaluation Understudy (BLEU), BERTScore, etc.). Framework or platforms that aid in rapid assessments of agentic AI are of interest. While these metrics are commonly associated with evaluating non-agentic AI systems (e.g., static models for classification or prediction), they are equally applicable to agentic AI systems in specific contexts.

4.8. Optimization for Tactical and Low-SWaP-C Resources (i.e., methods, tools, frameworks) – hyperparameter tuning, model sparsity, quantization, pruning, ensemble methods, etc.

4.9. Collaborative Autonomous Systems

4.9.1. Multiagent Coordination:

4.9.1.1. Multi-Agent Systems (MAS)

4.9.1.2. Decentralized decision-making frameworks that enable autonomous agents (ground, air, and sea) to collaborate without central control, adapting to dynamic environments.

4.9.1.3. Goal decomposition and task allocation among the swarm.

4.9.1.4. Distributed task allocation across multiple unmanned systems to maximize efficiency and effectiveness in real time missions.

4.9.1.5. Self-organizing systems that allow autonomous communications, information sharing, and synchronization of effort, even in GPS or communication denied environments.

5. Expectations and Exemplars: This paragraph lists some topics that would be of interest. Technologies from section 4 may be involved in one of the exemplars below. It is intended to give respondents an idea of areas of interest and expectations regarding the level of sophistication. USSOCOM wants to restrict this TE event to new and novel truly agentic based (especially multi-agent) solutions. The intent of this TE is to discover new solutions that USSOCOM has not seen before. Therefore, USSOCOM is open to respondent input beyond what is listed here. Participants will be screened prior to participation to ensure their solutions fit the intent of this TE.

5.1. Cloud and Haze Detection – The ATR may not be programmed to detect cloud obscuration (total or periodic) or haze that can significantly affect ATR performance. There are cloud and haze analogies when dealing with ATRs in other domains or phenomenology. A multi-agent system could detect these conditions and adjust altitude (maneuver), slant angle, gimbal scan pattern (e.g., look back after a small, fair-weather cumulus cloud obscuration), etc. Mission Autonomy (MA) understands sensor ball capabilities and adjusts gimbal scan and maneuver appropriately based on intended targets and biome.

5.2. Reporting Quality – A UxS reporting nothing may be vastly different than a UxS reporting the absence of targets with x confidence. Operator interest: Was the UxS able to clearly observe the Area of Interest (AOI), and can it confirm that it saw nothing of interest? This is much different than a UxS reporting nothing. There may have been total cloud obscuration, and the ATR & platform are oblivious to it when they couldn’t even observe the background where the targets would be. The system should be able to confirm the absence of targets on the background (e.g., ATR can confirm no obscuration, complete coverage of the AOI, and lack of identifiable targets in the AOI). If it cannot detect clouds, haze, or other obscuration, then this assessment cannot be made.

5.2.1. Exemplar #1: An agentic system may have a cloud detection algorithm separate from the ATR, or the ATR itself has this ability. A weather agent in concert with the Mission Autonomy (controlling maneuver and gimbal) should assess if haze or clouds are affecting or will affect ATR performance at x altitude, solar illumination geometry, etc. A “report agent” may augment reporting by issuing a J17.0-Weather Over Target in addition to a J3.3 message.

5.2.2. Exemplar #2: A “report agent” may monitor ATR detection output, MA behavior, Emission Control (EMCON) state, Commander’s Critical Information Requirements (CCIR)/ Priority Intelligence Requirements (PIR) updates, etc. to determine how and what to report [to whom] and in what message format(s). A sophisticated, multi-object ATR model may report certain High-Value Target (HVT) to strategic level command(s) and tactical threats to the tactical level teams. In a target rich environment with limited bandwidth/throughput, reporting everything to everyone may not be desired. The Information Exchange Requirements (IER) may call for reporting to different echelons with different radios. This logic may be onboard a UxS and/or offboard.

5.3. Super-resolution – Super-resolution has the potential for significantly improving ATR performance, but this processing comes at a cost in compute resources and latency. In a low-SWaP-C environment, compute, memory, and storage resources are limited and must be used wisely. If there was an agent that decides whether super-resolution is needed in a particular context, then it could be added to the processing workflow for a particular mission, which may be a tradeoff in frames per second processing, but the benefits may outweigh the cons. This decision may involve variables such as mathematical camera model, ATR capabilities in a particular biome, target(s) of interest for this mission, dynamic weather conditions, etc. to prognose how the setup should perform during the mission. This may require data from other agents (e.g., weather agent, MA agent, threat agent, etc.).

5.4. Sensor Geometry – Understand optimum slant angle or geometry of sensor or other sensor aspects to optimize sensor effectiveness. For example, a polarimetric camera can significantly help with target discrimination, but it is sensitive to sun geometry with respect to the target. Not every MA module will or should understand this geometry problem for a specific sensor phenomenology. Thus, a multi-agent system with a “sensor expert agent” can coordinate with the MA agent to best employ the platform to execute the mission.

5.5. Multi-sensor – There may be other sensor types and multiple sensors onboard (or offboard) to enable data fusion to increase object detection confidence. A “dynamic targeting process agent” could understand the authorities allocated to the machine (that may change with the dynamic environment) and make decisions WRT Positive Identification (PID), Combat Identification (CID), Target Engagement Authority (TEA), Collateral Damage Estimation (CDE), etc. This agent may require interaction with a “sensor expert agent” that understands what sensor(s) should be combined (and allocated processing to limited onboard resources). Another agent may be responsible for assessing confidence level of detection and identification/classification by understanding the ATR capabilities in a given situation.

5.6. Optimization – Account for target kinematics (e.g., car can’t be in swamp, so gimbal scan can skip that area or deprioritize it). Apply knowledge of land use, slope, maneuverability, etc. to adjust gimbal scan. One “kinematics agent” may coordinate with the “gimbal scan agent” and the “MA agent” to optimize gimbal scan patterns.

5.7. Authorities Construct – There are various authorities that may be allocated to the autonomous system under various conditions that are dynamic. These authorities include Positive Identification (PID), Combat Identification (CID), Target Engagement Authority (TEA), etc. These are likely different for different effectors. For example, the autonomous system may be authorized PID, CID, and TEA for Electronic Warfare (EW) effectors but not kinetic. An Agentic AI system could consider overlapping constraints, authorities, etc. to assess correct behavior.

5.8. Priorities of Constraints – There are numerous constraints from various data sources on the behavior of an autonomous agent. Some include Rules of Engagement (ROE), Weather, Weapons Control Status (WCS), Force Protection Condition (FPCON), Emission Control (EMCON) state, etc. These constraints overlap and sometimes include conflicting guidance. There may be different agents dealing with various constraints. In summary, some constraints should have priority over others, but each agent working on its constraints needs to be combined with the collective understanding of constraints. This may involve interaction among these agents or maybe another agent considering all their inputs. Constraints include:

· Rules of Engagement (ROE)

· Airspace Coordinating Measures (ACM)

· Fire Support Coordination Measures (FSCM)

· Weapons Control Status (WCS)

· Force Protection Condition (FPCON)

· Weather

· Emission control (EMCON) state

· Joint Restricted Frequency List (JRFL)

· Collateral Damage Estimates (CDE) thresholds

· Etc.

5.9. Rules Language – ROE and other constraints derive from various sources such as U.S. Code (e.g., Title 10 §130i for Counter-UAS), Policy memos, Chairman of the Joint Chiefs of Staff Instruction (CJCSI) 3121.01, Commander’s guidance, DoD Directive 3000.09, Special Instructions (SPINS), Executive Order (EXORD), Operation Order (OPORD), Operation Plan (OPLAN), ROE Serials, etc.). These data (or sanitized versions of it) must be represented in a way that can be read, parsed, and understood by agent(s) onboard the UxS (i.e., a “rules language”). Assuming the UxS has been allocated some authorities (e.g., EW effector employment under certain conditions), it must comply with its allocated authorities and various constraints such as ROE. Low-SWaP-C platforms do not have the resources (i.e., compute, memory, storage) to run a Large Language Model (LLM). Many may not be able to run a Small Language Model (SLM). Even with Natural Language Processing (NLP) and contextual understanding, implementing constraint enforcement via these mechanisms may not be allowed due to potential for misunderstanding leading to unforeseen or unallowed behavior. A traditional rules engine may be the most appropriate solution. Regardless, a rules language (possibly implemented in JavaScript Object Notation (JSON)) seems to be a valid approach for passing these constraints to a UxS. An onboard “rules agent” or “constraint agent” would receive, read, parse, understand, and enforce these rules by interacting with other agents onboard.

5.10. Override Conditions – There may be override conditions in certain cases (e.g., bingo fuel override [i.e., do not return to base] if it found a HVT and need to track it). These overrides relate back to constraint prioritization in a previous paragraph. There may be multiple agents working on their scope of tasks and behavior. These agents may derive conflicting behavior under certain conditions. When this occurs, they may need to be another agent that resolves these conflicts and authorizes override conditions on one or more constraints. EMCON state is another example that may be appropriate to override. A UxS may be in EMCON based on mission parameters and threat level. However, discovery of a CCIR may allow breaking EMCON.

5.11. Multi-tasking: Each UxS or swarm has a primary mission, but “everyone” on the battlefield is supposed to watch for Commander’s Critical Information Requirements (CCIR), Priority Intelligence Requirements (PIR), and HVT. This should not exclude UxS or intelligent agents running outside UxS. The ATR may be built for the primary mission, but, if resources allow, there could be another ATR (not necessarily Computer Vision-based) that looks for CCIR, PIR, & HVT. Another agent may receive data identifying the CCIR, PIR, & HVT (as they change) and selecting appropriate ATR(s) to find them.

5.12. Understand Battlespace Management (BM): Agents, whether onboard or offboard a UxS, need to understand doctrine WRT deliberate and dynamic targeting (and other mission area doctrine and TTPs as applicable such as resupply and medevac). Rather than translating BM corridors to geofences, agents should be able to receive, parse, and understand BM constraints such as Airspace Coordinating Measures (ACM). Referencing Joint Pub 3-52 (Joint Airspace Control), ACMs such as a No Fire Area (NFA) cannot be satisfied with simple geofences. An NFA is “an area designated by the appropriate commander into which fires, or their effects are prohibited”. The UxS may transit the NFA, but any “kinetic use agent” must understand that any use of an onboard effector cannot allow effects of fires in the NFA.

5.13. Semantics and Ontology: AI (except LLMs, when given access and context) do not understand and derive context from common data models (schema) that are missing semantics and ontology. Low-SWaP-C platforms and ATR need to report object identification or classification in some ontology (e.g., JIKO or DICO). For example, a “tank” can be a vessel for various liquids or a terrestrial battlefield weapon. The distinction is critical for treat awareness, etc. From the rules language mentioned elsewhere to messages received or sent by AI-enabled systems, Data Ontology Work (DOW) needs to mature its message definitions to include a knowledge representation amenable to intelligent agents.

5.14. Single Platform and Among Platforms – agents onboard a single UxS platform or agents collaborating between multiple platforms are critical to realize autonomous systems behavior. Not only onboard, but offboard AI-enabled systems doing non-UxS stuff or involved in UxS planning must interface Mission Command and other systems to receive mission tasking, commander’s intent, and various messages containing constraints. Without these System of Systems (SoS) interaction, agents cannot receive and send data to achieve critical mission threads.

5.15. Swarm Comms/Collaborative Autonomy – One agent may be responsible for single UxS behavior (e.g., MA agent), but a different agent (coordinating with MA) may be responsible for negotiating with a swarm for own-platform tasking. A “comm agent” may affect the radio available for swarm comms. It may have to coordinate with a “threat agent” to make best comm path decisions.

5.16. Understand Effector Capabilities (in context) and Related ROE – Not only must an autonomous system (and offboard decision support systems) receive ROE, but they must also understand how one or more effectors relate to the ROE. With the overlapping rules/constraints and their priorities, effector employment may change within the dynamic environment. For example, a UxS may be allocated authority to employ an EW effector in certain conditions. However, subsequent employment may change after a first wave of attack. That is, the effector may be authorized to use higher energy levels to defeat the 2nd wave of attack. This may involve Collateral Damage Estimation (CDE) by a human (and received by the agent) or by a “CDE agent” that affects effector employment within the ROE. These change by effector type, so the construct must account for this complexity. Can the EW effector be used for this frequency? Not if it’s on the JRFL. An authorized agent controlling an EW effector must receive JRFL updates (or a sanitized version of it) to ensure these spectral constraints are implemented. Similar constraints apply to directed energy and kinetic effectors.

5.17. Example Agents – Threat, weather (Wx), maneuver, gimbal scan control, comm control & data priority, cognitive radio/ Software-Defined Wide Area Network (SD-WAN), MA, planner, scheduler, constraint resolver, resource monitor and planner, dimensionality reducer (in context), PID/CID and confidence level evaluator (how can I achieve ID threshold), fault isolation and troubleshooting, capes/status (Fully Mission Capable (FMC)/Partially Mission Capable (PMC)/Not Mission Capable (NMC) reporting, etc. These aspects may be an intelligent agent that handles related information and behavior.

5.18. Anomaly Detection and Prosecution – AI-enabled systems (especially those that employ behavior trees) will experience anomalies. The primary mission may be finding a ship class on the open ocean and then tracking it. However, the ATR may not be trained to find “other interesting things in the video feed.” If there is a plume of smoke near the edge of the camera’s field of view (FOV), what should or could be done? The smoke may be obscuring the exact class of ship the UxS is supposed to find. If it does not investigate the plume, additional context cannot be gained. The smoke plume may be a distraction or decoy, so the decision space is complex. Fault conditions are another anomaly that an autonomous system may not be programmed to handle. If a sensor ball has two cameras and one fails, what should the autonomous system do? If a set of variables are realized that do not conform to a branch on the behavior tree, what should the autonomous system do?

5.19. Decision Support Systems – There are numerous examples that can be demonstrated fusing intelligence information and providing automated notifications or weighted decision support CoAs. Similarly, fused intelligence can be used to redirect ongoing operations or provide alternate weighted CoAs.

6. Security/Classification Requirements: Respondents shall not submit classified information in the technology experimentation nominations.

7. Safety Requirements: All respondents shall review the TE Safety Guide (RFI Notice Attachment 3). Those respondents who are invited to demonstrate their technologies must complete a Deliberate Risk Assessment Worksheet (Department of Defense Form 2977) (RFI Notice Attachment 4) in accordance with MIL-STD-882E and the Department of the Army Techniques Publication No. 5-19 (ATP 5-19). Risk assessments shall be emailed directly to the tech_exp@USSOCOM.mil by 4 March 2026. Respondents should include instructions that describe the safe operation of the device nominated for the experiment.

8. Frequency Requirements: If your experiment has any device radiating on a given frequency or frequency band, you must have prior approval to transmit on that frequency. Prior approval includes compliance with Federal Communications Commission (FCC) Title 47, Part 15, or a Special Temporary Authority (STA) from the FCC. You must have National Telecommunications and Information Administration (NTIA) frequency approval if your experiment includes Government-owned equipment, and you will be operating within a Federal Band. Respondents are advised not to wait for confirmation of selection/invitation to the event before requesting a STA from the FCC. Your authority to radiate must be emailed directly to tech_exp@USSOCOM.mil by 6 April 2026. All frequency questions shall be directed to the USSOCOM Technical Experimentation team at the email above.

9. Human Subjects. It is not anticipated that activities being conducted in this TE event will require the use of research or experimentation involving human subjects. Technology experiment submissions will be reviewed for potential research or experimentation involving human subjects. Any submission that is determined to potentially include research or experimentation involving human subjects will be required to adhere to DoD Instruction 3216.02 “Protection of Human Subjects and Adherence to Ethical Standards in DoD Supported Research” and ensure appropriate Institutional Review Board and DoD Human Research Protections Office approvals prior to conducting those activities.

10. Other Special Requirements: DO NOT SUBMIT PROPOSALS OR MARKETING DEMONSTRATIONS. SUBMIT TECHNOLOGY EXPERIMENTATION NOMINATIONS ONLY. EXPERIMENTATION NOMINATION SUBMITTALS FOR THIS RFI WILL ONLY BE ACCEPTED UNTIL THE CLOSING DATE OF 12 January 2026, 12:00 Noon Eastern Standard Time (EST), for the requirements stated above. No contracts will be awarded based solely on this announcement or any subsequent supplemental RFI announcements. All responses must be agentic; non-agentic solutions will be excluded.

C. SUBMISSION INSTRUCTIONS:

Technology experimentation nominations shall be submitted electronically by creating a Scout Card in the Vulcan system at:vulcan-sof.com by the respective deadline. Scout Cards should be created within the TE 26-2 Technical Experimentation: Agentic AI Integration collection. The Vulcan system is best accessed by using Firefox or Google Chrome browsers. Associated technology experiments with distinctly different uses or applications should have a separate nomination submitted by each respondent. USSOCOM personnel will review submissions to determine whether an experiment submission will be accepted for invitation to attend the TE event.

A complete submission consists of:

· Creating a Scout Card in the Vulcan system as described above.

· Completely filling in the Supplemental Information Electronic form via https://forms.osi.apps.mil/r/bLXTv0S8hQ?origin=lprLink. The Electronic Supplemental Information form is a MANDATORY requirement for ALL respondents. Failure to fill out the electronic form via the link provided is grounds for immediate rejection for this event.

· Devices with radio frequency emissions must state the intended frequency or frequencies used by the device(s).

· An FCC STA, Experimental License, or NTIA document (for developmental radio frequency emitting devices). If neither is available at the time of submission, provide status of your FCC/NTIA request. Those only using the ISM Band must state the exact frequency range and must transmit at less than 1watt.

· ISM band rules require devices to operate without causing harmful interference and to accept any interference from other devices. In the United States, these rules are governed by the FCC, with Part 18 for ISM equipment and Part 15 for unlicensed communication devices that share ISM frequencies.

· Instructions on how to safely use the technology (as needed).

· If applicable, a picture of the device with a short description of the size (shows the dimensions or places the device next to a ruler, currency, or man-sized object for comparison).

· A signed Photo Release (RFI Attachment 2).

Selected respondents will be invited to participate in USSOCOM experiments. USSOCOM shall provide venues, supporting infrastructure, and assessment (operational and technical, based on availability of resources and written request as discussed above) personnel at no cost to invited respondent(s). All respondents’ submission costs, travel costs, technology experiments, and experimentation associated costs will be at the respondents’ expense. The TE venue will only provide basic access to training areas or ranges (if approved and applicable) to conduct experiments, a facility to connect to the internet, basic venue infrastructure including frequency coordination/deconfliction, and shore power. Invited respondents must be prepared to be self-sufficient during the execution of their experiments and not dependent on venue resources.

Time and space will be made available for technology developers to conduct real-time modifications and updates to technologies. Technology developers are advised to bring all tools and equipment necessary to present/operate their technology at the event.

D. BASIS FOR SELECTION TO PARTICIPATE:

Selection of respondents to participate shall be based on the extent to which the technology represents a potential capability increase to SOF.

Other considerations include:

· Technical maturity

· Relevance of or adaptability to military operations/missions

· Relevance to current operational needs

· Relevance to Event Focus Area (agentic AI) E. ADDITIONAL INFORMATION: All efforts shall be made to protect proprietary information that is clearly marked in writing. Technology Developers shall not display any proprietary or classified information at the event. Lessons learned by USSOCOM from these experiments may be broadly disseminated, but only within the Government. If selected for participation in Technical Experimentation, developers may be requested to provide additional information that will be used in preparation for the experiments.

Future TE events with TENTATIVE themes/focus areas, locations, ANTICIPATED dates and RFI open periods are as follows:

Dates RFI Open Period

Technology developers invited to this event are encouraged to contact each other using social media through USSOCOM Technical Experimentation. While USSOCOM Technical Experimentation has access to several social media fora, the preferred collaboration social media link is LinkedIn www.tinyurl.com/LinkedIn-USSOCOMTE.

F. USE OF INFORMATION: The purpose of this notice is to gain information leading to Government/Industry collaboration for development of USSOCOM technology capabilities and to assist in accelerating the delivery of these capabilities to the SOF warrior. All proprietary information contained in the submission and technology experimentation shall be appropriately marked. The Government will not use proprietary information submitted from any one firm to establish future capability and requirements. During the event USSOCOM may broadcast technology experiments via video teleconference back to USSOCOM and/or their operational components. Respondents to this RFI shall review and submit the photography approval form at Attachment 2.

G. SPECIAL NOTICE:

1. Federally Funded Research and Development Centers (FFRDCs) or contractor consultant/advisors to the Government will review and provide support during evaluation of submittals. When appropriate, non-Government advisors may be used to objectively review a particular functional area and provide comments and recommendations to the Government. All advisors shall comply with procurement integrity laws and shall sign non-disclosure statements. The Government shall take into consideration requirements for avoiding conflicts of interest and ensure advisors comply with safeguarding proprietary data. Submission in response to this RFI constitutes approval to release the submittal to approved Government support contractors.

2. There will be foreign attendees (military, government, or industry partners), who are interested in the capabilities being demonstrated at the TE event. Respondents are ultimately responsible for complying with all International Trafficking in Arms Regulations (ITAR) /Export Administration Regulations (EAR) requirements associated with their equipment. Before declaring their ITAR/EAR status, respondents to this RFI should verify the ITAR/EAR status of their technology by visiting the US Department of State Directorate of Defense Trade Controls (ITAR) website at:

(https://www.pmddtc.state.gov/ddtc_public) and the US Department of Commerce Bureau of Industry and Security website (https://www.bis.doc.gov/) for more information on Export Administration Regulations (EAR). USSOCOM event organizers will restrict access as necessary to assist in protecting ITAR/EAR related technology demonstrations.

H. Per Federal Acquisition Regulation (FAR) 52.215-3, Request for Information or Solicitation for Planning Purposes (Oct 1997):

1. The Government does not intend to award a contract based on this RFI notice or to pay for the information.

2. Although "proposal" and "respondent" are used in this RFI, your responses will be treated as information only. It shall not be used as a proposal.

3. In accordance with FAR 15.209(c), the purpose of this RFI is to solicit technology experimentation candidates from research and development organizations, private industry, and academia for inclusion in future experimentation events coordinated by USSOCOM.

Contracting Office Address:

7701 Tampa Point Blvd MacDill AFB, Florida 33621-5323 Primary Point of Contact: tech_exp@USSOCOM.mil

(813) 826-4646

Endnotes 7 | T E R F I

UNCLASSIFIED

1 | T E 2 6 - 2 R F I UNCLASSIFIED

12 | T E 2 6 - 2 R F I