Information & Communications Technology

It has been credibly hypothesized - but not proven - that a fault-tolerant quantum computer of sufficient size would revolutionize multiple commercial industries and scientific disciplines. Quantum computers are shown to have transformative potential for critical problems facing the United States, it is in the Government's interest to foster and accelerate commercial progress towards a truly useful, "utility-scale" quantum computer. Initiated under Alternative Computing to both reduce strategic risk and realize transformative opportunity, the US2QC thrust will (1) evaluate disruptive designs for utility-scale, fault-tolerant quantum computers, specifically, systems that can be constructed in less than 10 years; (2) demonstrate each of the enabling sub-systems and components for these designs; and (3) construct a prototype fault-tolerant quantum computer that demonstrates that utility-scale design is viable. This program will continue as the Quantum Benchmarking Initiative (QBI) in PE 0602716E, Project ELT-02 in FY 2025, and in PE 0602025E, Project MQB-01 in FY 2026.

The Artificial Intelligence Quantified (AIQ) program is developing technologies for assessing the capabilities of generative artificial intelligence (genAI) to enable mathematical guarantees on performance. Current state-of-the-art methods for genAI quantification and assessment are ad hoc, deal with only the simplest of capabilities, and are not properly grounded in a rigorous theory. The AIQ approach combines mathematical methods with advances in measurement and modeling to enable rigorous quantification of genAI capabilities. The program will address three capability levels: 1) the specific problem level, which addresses quantification for a given problem; 2) the classes of problem level, which address quantification of a given problem class; and 3) the natural class level which addresses the reverse question of the class of problems quantifiable for a given genAI architecture. If successful, AIQ will provide guarantees on the performance of genAI technologies. Beginning in FY 2026, this program will be funded in PE 0602025E, Project MSL-05.

The Anticipatory and Adaptive Anti-money Laundering (A3ML) program aims to develop graph, search-based algorithms that enable anticipation of and adaptation to money laundering tactics, techniques, and procedures (TTPs) in order to prevent money laundering in the global financial system. Today, anti-money laundering (AML) practitioners accumulate evidence of money laundering largely via manual searches of myriad commercial and financial intelligence (FININT) databases and rule-based AML systems that lack flexibility to adapt to changing TTPs. Because of the static nature of current AML processes and the limited capabilities of AML tools, novel money laundering schemes can go on for years. To address this challenge, A3ML will develop and apply graph search and subgraph isomorphism techniques that reference high-level models of money laundering to extract specific money laundering TTPs from heterogeneous transaction data and to anticipate and adapt to new money laundering TTPs in a timely fashion. If successful, A3ML technologies and tools would enhance AML practices across U.S. Government (USG) and the private sector. Beginning in FY 2026, this program will be funded in PE 0602025E, Project MSL-05.

The Kallisti program seeks to develop new capabilities in theory of mind for behaviors in independent agents. Actions taken by independent agents are determined by their belief about the state of their environment, the risks and rewards associated with these beliefs, and the strategy they use to react to the risks or rewards they believe exist in their environment. This program will develop an algorithmic theory of mind that enables behaviors in independent agents by inferring and understanding the belief states that will deter or compel distributions of actions in an independent agent. Motivated by the Theory of Mind demonstrated in the Strategic Chaos Engine for Planning, Tactics, Experimentation and Resiliency (SCEPTER) program (budgeted in PE 0603760E, Project CCC-02), Kallisti will further develop this theory of mind to allow projection of an agent's future behavior based on the agent's perceived strategy, enabling influence over the agent. The program seeks not only to understand an agent's current strategy but also to find a decomposed version of the strategy into relevant basis vectors to track strategy changes under non-stationary assumptions. The primary goal is to utilize the algorithmic theory of mind to deter unwanted actions and compel more favorable actions in an agent's action space. Technology developed under this program will transition to the Services. Beginning in FY 2026, this program will be funded in PE 0602025E, Project MSL-05.

The Securing Artificial Intelligence for Battlefield Effective Robustness (SABER) program aims to advance understanding of the operational vulnerabilities of the military use of artificial intelligence (AI) enabled systems in battlefield environments and enhance the robustness of AI-enabled systems against real-world adversarial attacks. AI-enabled systems have been shown to be vulnerable to adversarial AI attacks, but most attacks studied to date are not physically realizable or assume the adversary already has some level of system access. SABER anticipates adversaries will develop combined AI-cyber-physical attacks and formalize these attacks in tactics, techniques, and procedures (TTPs). To address such capable AI adversaries, SABER will evaluate AI-enabled military systems in realistic operational settings to quantify the risks of deploying AI-enabled systems on the battlefield. Simultaneously, SABER will develop technologies and TTPs for continuous monitoring, integration, and deployment of counter-AI techniques to enhance the robustness of AI-enabled systems. SABER technologies will enable the formation of Department of Defense (DoD) AI red teams that continuously integrate and employ emerging counter-AI techniques and tools, establishing a sustainable model for operational AI red teaming. If successful, SABER will empower military AI red teams to confidently assess and secure U.S. AI-enabled battlefield systems against attack by sophisticated adversaries. Beginning in FY 2026, this program will be funded in PE 0602025E, Project MSL-05.

Building on insights developed from the Open Price Exploration for National security (OPEN) program (budgeted in PE 0602303E, Project IT-04), the National Security Economic Theory (NASCENT) program will perpetuate a policy and market feedback loop between the United States Government (USG) and private-sector executives to support more effective geoeconomic actions with precisely defined positive national security outcomes. Current approaches to geoeconomics are based largely on qualitative ideas, drawn primarily from international relations and legal backgrounds, and are often non-falsifiable with limited scope. Furthermore, the lack of a body that facilitates neutral and open discussion and iterative development of geoeconomic theory has hindered innovation and renders the development of theory and tools of economic warfare cost ineffective. NASCENT will overcome these challenges by creating a membership-based forum where the USG, academia, and the private sector can collaboratively and iteratively design and rigorously evaluate geoeconomic mechanisms. NASCENT will combine techniques from economic mechanism design and experimental economics to work backwards from precisely described national security outcomes, elicited from across the USG, to understand what economic incentives and mechanisms enable those outcomes. Beginning in FY 2026, this program will be funded in PE 0602025E, Project MSL-05.

The Accelerating Artificial Intelligence (AAI) program seeks to go beyond commercially-driven advances in artificial intelligence (AI) and to address important national security challenge applications. Trustworthy AI, which is AI that is safe, reliable, accurate, explainable, and resilient to attacks, is a major focus. Technical challenges include robustness of AI systems in novel, uncertain, and/or unanticipated situations; efficiency and timeliness of AI development, test, evaluation, approval, and certification processes; and identification of tasks or sub-tasks for which greater automation through the use of AI/machine learning (ML) is appropriate. Approaches to addressing these challenges will leverage recent advances in transfer learning, causal reasoning, reinforcement learning, generative AI, large pre-trained models, and large language models. If successful, AAI will significantly accelerate AI innovation in many important Department of Defense (DoD) domains while also reducing the time and cost needed to transition and deploy new AI technologies.

The Automating Scientific Knowledge Extraction and Modeling (ASKEM) program developed technologies and tools for the agile creation, sustainment, and enhancement of complex models and simulators to enable knowledge extraction and data-informed decision making in diverse scientific domains and military missions. Current modeling and simulation pipelines do not maintain the relevant inputs, assumptions, and modeling choices made during development. Rapidly changing knowledge, semantically-opaque models, and black-box simulators make pipelined development nearly impossible. ASKEM enables a new paradigm for scientific modeling analogous to the transition in software development from the lengthy waterfall model to agile, continual Development and Operations. ASKEM modeling automation tools 1) extract model components from documents and code, while abstracting implementation details like math framework, language, and platform; 2) compose distinct model and simulator components; and 3) integrate all elements and processes in an extensible workbench that addresses the entire modeling and simulation lifecycle. ASKEM tools enable experts to maintain, reuse, and adapt large collections of heterogeneous data, knowledge, and models with traceability across knowledge sources, model assumptions, and model fitness and thereby bring agile, pipelined development to modeling and simulation.

The Automated Rapid Certification Of Software (ARCOS) program developed technologies that automate the capture and evaluation of software assurance evidence which enables certifiers to assess system risks earlier in the process and to commit to engineering decisions more rapidly and safely. Historically, software certification practices did not scale with the extent, complexity, and interconnection of software being developed by the Department of Defense (DoD), so certification had become a bottleneck to new system deployment. ARCOS technologies address DoD software system certification time and cost. Specifically, ARCOS technology automatically and interactively generates strong assurance arguments that incorporate supporting evidence for certification criteria. ARCOS also developed techniques to compose assurance arguments for pre-evaluated components into consolidated assurance arguments for new systems incorporating those components.

The Scientific Feasibility (SciFy) program is developing computational methods to measure the feasibility of claims to enable accurate assessments of scientific content. Automated scientific content generation, via rapidly improving large pre-trained models, has the potential to disrupt the U.S. technology base in times of crisis and to distort the global race for technological dominance in key areas. Similarly, false capability claims can have significant negative implications for national security and international relations. To address these threats, SciFy will focus on methods for assessing the scientific feasibility of claims by using automated reasoning to decompose claims into constituent, verifiable parts. Assessing each component will involve referencing existing technological advancements, foundational scientific principles, data, software, models, simulation results, and industry standards or benchmarks. SciFy will create methods that go beyond automated fact-checking by addressing complex component interactions and operational constraints as well as evaluating logical consistency, system integration, and compatibility considerations. If successful, SciFy will enable the U.S. to reliably determine whether claimed scientific and technological capabilities are practical and realistic when considered as a whole, even when theoretically possible in parts. This applied research funding supports applied research efforts that are part of the larger basic research SciFy activity under PE 0601101E, Project CCS-02. Beginning in FY 2026, this program will be funded in PE 0602025E, Project MSL-05 and PE 0601122E, Project EMR-01.

The Assured Autonomy program developed rigorous design and analysis technologies for continual assurance of learning-enabled autonomous systems to enhance system safety in uncertain environments. Typically, the state of the art for test, evaluation, verification, and validation is only applicable to non-learning systems operating in well-characterized environments. As a result, autonomous systems enabled by machine learning (e.g., deep neural nets for perception, reinforcement learning for control policies, and online model learning) can be challenging with respect to rigorous safety assurance. Assured Autonomy developed new techniques for modeling and system design, formal verification, simulation-based testing, and safety-assured learning to provide continual assurance of learning-enabled autonomous systems. The technologies developed in Assured Autonomy enable the Department of Defense (DoD) to more rapidly and efficiently deploy learning-enabled autonomous systems that can be trusted to operate safely in uncertain environments.

The Transfer from Imprecise and Abstract Models to Autonomous Technologies (TIAMAT) program is developing techniques to robustly transfer learned autonomy from fast abstract simulations to autonomous platforms in real-world environments. The autonomy levels of unmanned systems of today are limited because the modeling and simulation (M&S) training environments do not account for the data domain shift common when translating M&S outcomes to the real world, a phenomenon sometimes referred to as the sim2real gap. The TIAMAT approach integrates symbolic structures with neural structures to transfer learned autonomy more realistically and robustly. TIAMAT will enable the use of fast abstract simulations by anchoring the learning and transfer of autonomy on semantically consistent components shared across simulations and real environments, so-called "semantic anchors". For TIAMAT, semantically consistent components of particular importance include militarily relevant phenomena that remain consistent in the source and target environments, such as mission objectives, special instructions, subject matter expert guidance, rules of engagement, and the laws of physics. Autonomy transfer using TIAMAT methods and techniques will reduce the complexity of the autonomy learning and transfer problems to the comparatively simpler points of reference in the anchored representation. If successful, TIAMAT transfer of M&S-based learning will enable more rapid and robust training and deployment of autonomous systems at higher levels of autonomy. Beginning in FY 2026, this program will be funded in PE 0602025E, Project MSL-05.

The Assured Neuro Symbolic Learning and Reasoning (ANSR) program is developing new hybrid artificial intelligence (AI) algorithms that deeply integrate symbolic reasoning with data driven learning to create trustworthy AI-based systems. For purposes of this program, an AI-based system is considered trustworthy if it is: (a) robust to domain-informed and adversarial perturbations; (b) supported by an assurance framework that creates and analyzes heterogenous evidence towards safety and risk assessments; and (c) predictable with respect to some specification and model of fitness. ANSR develops hybrid AI algorithms for which it is possible to develop evidence-based techniques that support confident assurance judgments. The key idea is to interleave symbolic and neural representations in hybrid AI algorithms that can acquire symbolic knowledge through learning and perform symbolic reasoning at scale to deliver robust inference, generalize to new situations, and provide evidence for assurance and trust. ANSR technologies will be demonstrated and evaluated on Department of Defense (DoD) use cases, such as autonomy, where trustworthiness is essential. Beginning in FY 2026, this program will be funded in PE 0602025E, Project MSL-05.

The Learning Introspective Control (LINC) program is developing machine introspection and learning technologies to characterize a modified or damaged military platform from its behavior and update the control law to maintain stability and control. The current approach to handling platform modification or damage places the burden of recovery and control on the operator, whether the operator is human or an autonomous controller. In contrast, a platform equipped with LINC technology would continually compare the real-time behavior of the platform as measured by on-board sensors with a learned model, determine if the current observed behavior of the platform differs from that model in ways that might compromise stability and control, and implement an updated control law when required. The LINC capability would aid operators in maintaining effective control of military platforms that suffer damage or have been modified in the field to address emergent requirements identified during operations. Beginning in FY 2026, this program will be funded in PE 0602025E, Project MSL-05.

The Open Price Exploration for National security (OPEN) program aims to increase supply chain resilience and enable more efficient critical mineral markets by leveraging advances in artificial intelligence (AI) prediction and forecasting to increase structural price, supply, and demand transparency. OPEN will construct structural price predictions from fundamental and observable critical mineral input costs and increase the accuracy and precision of supply and demand forecasts by leveraging this structural price in conjunction with advances in AI and economic modeling. Today, critical mineral markets and supply chains are vulnerable. International supply shocks can lead to large and rapid critical mineral price spikes with immediate economic ramifications, and commodities purchase transactions (e.g., offtake agreements) are negotiated leveraging a mix of opaque and flawed pricing data. OPEN will leverage a decomposition of a critical mineral price into four components (input costs, supply/demand shocks, distortions due to noncompetitive behavior, and stochastic fluctuation) to construct transparent estimations of an approximate marginal cost for critical minerals indexed by time and geographic location, and will estimate supply and demand forecasts for critical minerals that take into account geopolitical factors, energy fluctuations, and technological innovations in recycling and supply chain management. Technology developed under this program will transition to applicable government entities and commercial partners. Beginning in FY 2026, this program will be funded in PE 0602025E, Project MSL-05.

The Artificial Intelligence Cyber Challenge (AIxCC) program is developing and demonstrating techniques for automated discovery and remediation of software vulnerabilities at speed and at scale to secure widely-used, critical code. Current automated vulnerability discovery and remediation tools are based on techniques such as fuzzing, logical reasoning, and genetic algorithms, but are limited in terms of effectiveness and user support. AIxCC leverages recent dramatic advances in artificial intelligence (AI) and machine learning, such as large pre-trained models and neurosymbolic AI, as the basis for new automated cyber security technologies and tools. AIxCC employs a contest model where teams use their automation and tooling to complete vulnerability discovery and remediation challenges. Performer teams are selected for the AIxCC competition based on their capability to leverage advances in AI to create usable, automated tools for vulnerability discovery and remediation and focus on tools suitable for broad deployment and applicable to critical infrastructure sectors. AIxCC competitors will train and develop their systems to find and fix vulnerabilities in widely used open-source software, focusing on software used in critical infrastructure. Each competitor system is evaluated on real-world critical infrastructure software suites and is scored based on their results in terms of absolute performance and performance relative to other competitor systems. Winning teams will receive cash awards. If successful, AIxCC will create novel AI-enabled cyber vulnerability remediation technology and tools for securing code at the scale and speed needed to defend U.S. critical infrastructure from cyber-attacks. Beginning in FY 2026, this program will be funded in PE 0602025E, Project MSL-05.

The Securing Information for Encrypted Verification and Evaluation (SIEVE) program developed technology to enable the creation of mathematically-verifiable, public statements derived from sensitive information that remains hidden. SIEVE produced advances in a cryptographic technique known as zero knowledge (ZK) proofs, which simultaneously enables mathematical verification of public statements while provably hiding the sensitive information from which the statement is derived. The advances produced by SIEVE made it operationally feasible to verify statements substantially more complex than the previous ZK state-of-the-art methods support, for example, statements about a software vulnerability that do not reveal details of how the vulnerability can be exploited.

The Fast Network Interface Cards (FastNICs) program created new networking technologies to accelerate the computation of distributed applications. As a result of incremental technology advances in networking and computing market silos, today's network and computing subsystems are badly out of balance with each other which produces a bottleneck at the network interface used to connect a machine to an external network and severely limits the input/output capability. FastNICs developed new input/output technologies based on more realistic models of complex, multiprocessor compute, interconnect, and memory subsystems. FastNICs enables a dramatic increase in computational throughput for distributed applications, such as training of machine learning systems.

The Program Analysis for Capability Excellence (PACE) program developed tools and techniques to autonomously identify adversary compromise of software, mitigate negative effects of adversary capabilities, and restore the integrity of compromised software. PACE enabled rapid, autonomous response to cyber attacks without using source code or requiring recompilation.

The Assured Micropatching (AMP) program developed technologies to enable the rapid production of targeted micropatches to repair legacy program binaries with strong guarantees. At present, the emergency patching of legacy software, even if all relevant information is available, creates too much uncertainty and takes far too long to validate, leaving critical systems with known flaws vulnerable to adversary attack. AMP created capabilities to analyze, modify, and fix legacy software in binary form even when the original source code and/or build process is not fully available. The AMP technical approach involved automatic discovery of known vulnerable components, goal-driven decompilation to isolate and analyze the vulnerable binary components, and minimal-change patching and recompilation to rebuild affected binaries with strong guarantees that the patch will not impair the functions of the system. The technologies developed by AMP enable cyber defenders to patch legacy binaries quickly and accurately in the deployed software systems upon which the Department of Defense (DoD) depends.

The Open, Programmable, Secure 5G (OPS-5G) program developed open source, 5G network software to ensure security and stimulate innovation in mobile wireless hardware. At the time of program inception, trends in mobile wireless technology development were unfavorable in that the U.S. and allies were increasingly dependent on proprietary technologies offered by foreign suppliers. OPS-5G developed standards-compliant software for 5G mobile wireless networks that is open source, programmable, and secure by design. The availability of open-source software for 5G has the additional benefit of opening the mobile wireless hardware market to new participants, stimulating innovation and competition. The OPS-5G program aimed to move the mobile wireless market away from its previous model of opaque, proprietary, and vertically-integrated technology provided by a small number of dominant foreign vendors to a more robust model with increased transparency and open-source technology created by a diverse ecosystem of U.S.-based academic and commercial software and hardware developers. OPS-5G was coordinated with existing open-source 5G efforts and U.S. Government, Department of Defense (DoD), and industry stakeholders.

The Verified Security and Performance Enhancement of Large Legacy Software (V-SPELLS) program is creating methods and tools to recover succinct models of domain data abstractions and logic from source code, add enhancements to the models, and convert them to performant new component implementations verified to be compatible and secure. The Department of Defense (DoD) has a critical need for replacing or reworking components of existing software with more secure and more performant code, including cases where a key performance or security benefit comes from moving parts of the software to new hardware, such as utilizing hardware accelerators, isolation enclaves, offload processors, and distributed computation. At present, the enhancement of legacy software components faces high risk that the new software will not be fully compatible with the existing, larger environment. Verified software is currently written from scratch, starting with a formal specification, rather than incrementally added to a system as provably compatible enhancements. V-SPELLS will address these problems by combining novel concepts in verified programming with recent developments in domain specific languages and systems architecture. V-SPELLS aims to enable piecewise, compatible-by-construction improvement of software components in legacy DoD systems, which will provide incremental software (re)engineering the benefits of formal software verification currently available only to clean-slate development efforts.

The Signature Management using Operational Knowledge and Environments (SMOKE) program is developing signature management technologies that generate evasive cyber infrastructure which minimizes signatures as a source of attribution. SMOKE technologies incorporate counter-attribution techniques into the design process, quantitatively measure attribution risk in real-time, and maintain evasiveness after infrastructure changes. SMOKE data-driven tools will automate the planning and execution of threat-emulated cyber infrastructure needed for network security assessments by red teams. SMOKE data-driven tools will automate the discovery of cyber threat infrastructure signatures. If successful, SMOKE prototypes will enable red teams to plan, build, and deploy cyber infrastructure that is informed by machine-readable signatures of sophisticated cyber threats.

The Reclaiming Bus-based Systems During Compromise (Red-C) program will develop and demonstrate novel approaches for resilient, self-healing, bus-based systems. The embedded computing systems common to modern vehicles and other cyber-physical systems typically use a bus-based architecture for communication among the various sensor, processor, actuator, and controller components organic to the platform. Due to the physical isolation of embedded computing systems, most standard buses currently in use were developed under an implicit assumption that all bus-connected components are secure and can be trusted. As a consequence, most bus-based systems are highly vulnerable in that compromise of any component on the bus can quickly spread to other bus-connected components. Red-C aims to address this critical vulnerability and create resilient, self-healing, bus-based systems through advances to instrumentation and response. For instrumentation, Red-C aims to enhance the individual components on the bus to enable them to function as sensors that collectively monitor the bus and peer bus components to detect a cyber-attack. Then, having identified that an attack is underway, Red-C-enhanced bus peers will collaborate on software patch creation to remediate bus vulnerabilities and enable recovery of bus functionality. Red-C will implement these bus resilience techniques as firmware updates that can be deployed to existing bus-based systems and components. Red-C technologies will mitigate bus vulnerabilities in military and civilian platforms and cyber-physical systems. Beginning in FY 2026, this program will be funded in PE 0602025E, Project MSL-04.

The Provably Weird Network Deployment and Detection (PWND²) program is developing formal models of emergent communication pathways to fundamentally improve how we deploy and detect resilient hidden networks in the real world. Authoritarian regimes are increasingly able to monitor and target internet communications, leaving many populations unable to communicate freely and safely. In response, the internet freedom and national security communities manually design hidden communication systems that use existing/available infrastructure and then validate these systems via empirical testing and analysis. These tests have difficulty capturing the range of real-world network environments and require infeasible amounts of testing to assess comprehensively at scale. PWND² is developing principled approaches for creating hidden communications systems with formal guarantees and for defending against use of these techniques by an adversary. If successful, PWND² technologies would enable the rapid and automatic generation and deployment of hidden communications systems with high assurance and strong guarantees. Beginning in FY 2026, this program will be funded in PE 0602025E, Project MSL-04.

The Pipelined Reasoning of Verifiers Enabling Robust Systems (PROVERS) program is creating scalable mathematically based technologies, tools, and practices to achieve continuous reasoning about complex systems that can support software development pipelines. These mathematically based techniques, or formal methods, enable rigorous modeling, reasoning, and proving of diverse properties of software code or design models, such as the absence of a specific type of defect or security vulnerability. PROVERS integrates formal methods into a modern incremental and iterative development process by running tools at each code commit and delivering results to developers when they can most effectively remediate discovered issues. PROVERS focuses on creating and sustaining a body of evidence that can co-evolve with the system under change to support continuous assessment and ensure that the system remains free of identified categories of defects and security vulnerabilities through its lifetime. Key PROVERS objectives include enabling proof maintenance and repair capabilities at a cost that is proportionate to code change; integration of formal methods with code, properties, and proofs in a single workflow that reduces human involvement; providing improved explanations to facilitate proof repair; and automating formal methods-based software analysis to support software developers that are not formal methods experts. PROVERS technologies will facilitate the agile development and continuous improvement of mission-critical software systems that meet the high security and quality standards required by the Department of Defense. Basic research for this program is funded in PE 0601101E, Project CCS-02. Beginning in FY 2026, this program will be funded in PE 0602025E, Project MSL-04.

The Intelligent Generation of Tools for Security (INGOTS) program is developing techniques to identify, triage, and assess exploitability of chainable vulnerabilities. Today, sophisticated cyber attacks link multiple vulnerabilities together into exploit chains that bypass software and hardware security measures to compromise critical, high-value systems. Accurately understanding risk is critical for both developers and defenders within cyberspace, but the metrics currently in use do not account for the multiple factors that differentiate an innocuous software flaw from a chainable vulnerability. INGOTS is developing semi-automated tools and techniques to characterize and measure the interdependent exploitability of vulnerabilities and a new vulnerability severity metrology that characterizes and measures interdependent exploitability. With the INGOTS vulnerability measurement capability, developers and defenders will improve software and hardware resiliency by rapidly identifying and prioritizing their most dangerous flaws. The INGOTS program is also funded in PE 0602716E, Project ELT-02. Beginning in FY 2026, this program will be funded in PE 0602025E, Project MSL-02 and MSL-04.

The Hardening Development Toolchains Against Emergent Execution Engines (HARDEN) program is developing techniques and tools to anticipate, isolate, and mitigate emergent system behaviors and thereby improve security of complex integrated software. Today's software development toolchains and testing methodologies provide very limited means for reasoning about adversarial reuse of code as written and designed. This limitation results in unwitting creation of stable, reliable patterns of emergent behaviors within systems that adversaries can reuse in attacks. The HARDEN approach to preventing adversarial code reuse is to create techniques, tools, metadata, and instrumentation for reasoning about emergent execution at all stages of the software development life cycle (SDLC) and to flag code segments and design patterns where there is high potential for adversarial reuse and emergent execution. The utility of HARDEN technologies will be assessed by applying them to integrated software systems and critical system elements, such as bootloaders. If successful, the technologies developed by HARDEN will facilitate efficient mitigation of complex code-reuse and emergent-execution vulnerabilities at early SDLC stages and provide the stronger roots-of-trust required by zero trust architectures and high assurance integrated military software systems. Beginning in FY 2026, this program will be funded in PE 0602025E, Project MSL-04.

The Business Process Logic (BPL) program is developing techniques to characterize and resolve vulnerabilities in business logic systems to protect and assure defense-critical workflows for government and business. Automated workflows written in business logic (BL) control much of the world's enterprises, from administration and operation of seaports to the assembly of weapons systems. Losses due to BL faults and vulnerabilities can range from annoyances to business-threatening outcomes. Therefore, it is important to identify and correct potentially problematic logic issues such as one-way actions or lost resources as early as possible. The BPL program is developing tools to extract workflow representations from BL and use those representations to identify, characterize, and mitigate faults and vulnerabilities in BL scripts and templates automatically. The technologies developed by BPL will enable increased assurance for manufacturing and assembly and greater efficiency for logistics and supply chain management. Beginning in FY 2026, this program will be funded in PE 0602025E, Project MSL-04.

The Cyber Agents for Security Testing and Learning Environments (CASTLE) program is developing an artificial intelligence (AI) toolkit to instantiate realistic network environments and train AI cyber agents to enable resilient network operations against advanced persistent threats (APTs). CASTLE formulates network hardening as a reinforcement learning (RL) problem and teaches RL agents to operate through the post-breach behavior of widely available penetration testing tools. Over progressive rounds of attack and defense, agents explore defensive actions to proactively stop ongoing attacks while maintaining operationally relevant workflows. Environments execute agents inside instrumented subnets that are deployed to live networks and simulate defensive actions that counter APT tools. Agent execution produces calibrated datasets for progressively improving simulations. The defensive AI cyber agents developed under CASTLE will provide the Department of Defense (DoD) with continual security assessments of critical networks and real-time response to cyber attacks. Beginning in FY 2026, this program will be funded in PE 0602025E, Project MSL-04.

The Constellation program is comprised of multiple projects that are developing technologies, capabilities, and prototype systems to enable full spectrum military cyberspace operations to defend the U.S. and deter, disrupt, and defeat adversary cyber actors. Technologies of interest include, but are not limited to, artificial intelligence, machine learning, and data science; resilient software, networking, and computing systems; formal methods and program analysis; data and information assurance; and cyber threat intelligence. High relevance is achieved through close coordination with U.S. cyber operators and the use of development, security, and operations (DevSecOps) and other collaborative development processes. High velocity is achieved through streamlined acquisition, assessment, approval, and deployment processes. Constellation development and deployment pipelines enable the rapid and continuous delivery of cyber technologies, capabilities, and prototype systems into operational use for the Department of Defense. The Constellation program is also funded in PE 0603760E, Project CCC-05 to facilitate rapid transition of cyber technologies and laboratory prototypes from applied research to operational prototypes. Beginning in FY 2026, this program will be funded in PE 0603469E, Project AET-03 and PE 0602025E, Project MSL-04.