The German Ministry of Defense (MoD), the Bundeswehr, and the defense technological and industrial base understand the importance of artificial intelligence (AI) in shaping the future strategic environment and the use of military power. Numerous projects have been launched, structures and processes are being reorganized, money has been earmarked, and training is underway or being readjusted.

Overall, however, Germany’s AI path remains murky as it is caught in a “master and servant” logic that will be painful to overcome. The metaphor describes the struggle to readjust Germany’s input-driven strategic culture, which puts greater emphasis on the socio-political acceptance and legitimization of military power than on the effects it can achieve. As a consequence, defense trumps offense with most of the current defense AI initiatives aiming at augmenting the survivability rather than the lethality of the Bundeswehr. Additionally, three decades of political neglect have pinned the Bundeswehr’s socio-technological imaginaries to what matters here and now. Force planners could (and would) not request, what they did not know. This triggered a kind of technology blindness that has—unintentionally—been reinforced by a technology agnostic approach to capability development that describes capability requirements in generic terms. Today, the Bundeswehr operates in a bifurcated world: the armed forces need to envision the future defense environment while procrastinating future concepts and projects into today’s procedures and processes—the master—to induce incremental change.

Against this background, Germany considers defense AI as a tool—the humble servant—subordinate to humans, who must always remain in the loop. Incrementalism dominates, which makes it difficult to assess what defense AI is expected to achieve and whether it delivers on this expectation. Although US ideas play a prominent role in the defense AI discourse, Germany’s strategic culture acts as a filter that tames the US emphasis on performance and lethality to make it more palatable for risk averse German decision-makers. As a result, most German defense AI projects focus on decision-making support and gradual improvements of other technologies in the fields of Command, Control, Computers, and Communications (C4) and Intelligence, Surveillance, Reconnaissance (ISR). In parallel, organizational adaptation is underway, but the MoD and the military services seem to operate at different levels of ambition and diverging speeds. It remains to be seen if the yet to be published defense AI implementation strategy will be able to remedy this shortfall.

Germany has stepped up investments in its digital infrastructure but slashed spending on defense research and technology (R&T). Based on a review of non-public budgets Germany currently spends around €50M per year on AI-related software development. As opaque as its spending is the Bundeswehr’s current fielding and operating of defense AI. An open-source intelligence system for crisis early warning, AI-based warning receivers for helicopters, and intelligent image processing for missiles feature among the more prominent, publicly known examples. Finally, defense AI affects military education and training. The Bundeswehr’s Command and Staff College is about to review its curriculum with the goal to incorporate AI elements as of 2024. In addition, the University of the Bundeswehr/Hamburg is setting up a new AI bachelor’s and master’s degree course. Individual services also explore opportunities for AI-enhanced simulation-based training. Moreover, different initiatives have been launched to train defense AI algorithms.

In sum, Germany has embarked on a defense AI journey, but substantial homework remains to be done. To this purpose the German MoD needs to be more precise about the future capability growth AI is expected to enable, the role of defense AI in Germany’s (non-existent) defense industrial policy, its international defense AI ambition, and the framework needed to certify, qualify, and admit future defense AI solutions.

1 Thinking About Defense AI

The German government’s 2018 AI strategy and its 2020 update (Federal Government 2018, 2020) describe Germany’s ambition and line of efforts to use AI to advance national and European competitiveness. But both documents remain silent on the use of AI for defense and security. The same is true for the 2018 Concept of the Bundeswehr, the 2021 strategic guidance of then Minister of Defense Kramp-Karrenbauer, and the 2021 coalition treaty. Only the 2019 concept paper on “AI for use in the area of responsibility of the Ministry of Defense” (BMVg 2019a) fills the void.

This is disenchanting, but it is not surprising, as Germany’s strategic culture tames the Bundeswehr’s technology appetite, creating tensions. The Bundeswehr recognizes that technology is changing the future battlefield. It also embraces allied concept ideas to signal its willingness to cooperate with partners. However, culture, the current organizational set up, and the lack of robust technology leadership pin the Bundeswehr down to the status quo. Defense AI thus has a hard time blossoming—in part also because its use is most often tied to large procurement projects that apply defense AI as part of broader functionalities. This makes it difficult to understand Germany’s overall defense AI ambition and the added value AI is expected to deliver.

1.1 Structural Pacifism Shapes Defense Technological Imaginaries

Germany’s security policy is characterized by a structural pacifism (Verbovszky 2024) in which the need to reconcile competing elements of Germany’s post-war security identity with a byzantine policy process leads to the prioritization of security policy conformity over effectiveness (performance).

Resulting from the trauma of WWII, the nascent Federal Republic of West Germany “reinvented itself” in opposition to its authoritarian past (Stengel 2020: 102). The formation of Germany’s post-war identity relied heavily on fantasy (Eberle 2019: 46)—a narrative scenario that promises the impossible fulfillment of a complete identity—and, more importantly, on negative contingency—the projection of catastrophe into the future, which is meant to be avoided. As Frank Biess (Biess 2019: 31) argues, negative contingency in the form of German Angst served to stabilize German democracy by emphasizing its fragility.

While conducive to internal stability, negative contingency retards German security policy. Reconciling the “Lessons of History” presents itself as the “right way” to do security policy, i.e., most conform with Germany’s post-war identity (Verbovszky 2024: 36). But reconciling competing interpretations of German post-war identity is made more difficult by the byzantine logic of German security policy decision-making. Myriad actors all come with their own political interests. Resolving them in a way that conforms with Germany’s post-war non-belligerent identity leads to a security policy dominated by inputs. Security policy decisions are done via “reverse consensus,” i.e., even before going into respective committees for deliberation they are designed to be consensus-capable in the final vote (Verbovszky 2024: 48).

Similar forces underpin Germany’s skepticism toward technological change. One tool for measuring the impact of cultural and political factors is the use of socio-technological imaginaries, i.e., “collectively held, institutionally stabilized, and publicly performed visions of desirable futures, animated by shared understandings of forms of social life and social order attainable through, and supportive of, advances in science and technology” (Jasanoff 2015: 4). They “play an important role in the development, assessment and regulation of cutting-edge technologies” (Burri 2015: 233). Comparative assessments across diverse technology areas like nanotechnology, space, and AI show, that these technologies are considered relevant for national competitiveness and to advance environmental assessments, whereas military applications receive scant attention. In addition, risk emanating from these technologies is used to express ever-present warnings justifying preemptive regulation as a preeminent political task, thus further embedding the political discourse in technology critical imaginaries (Federal Government 2018: 8; SPD/Bündnis90/Die Grünen/FDP 2021: 145; Burri 2015: 237–239; Kober and Schütz 2024).

1.2 Future Conflict Picture

Germany’s strategic culture shapes how the Bundeswehr thinks about the future. Its capstone document Future Operating Environment 2035 (BMVg 2019c) describes fast-paced adversarial action and the amalgamation of different forms of conflicts that evolve in complex and chaotic environments. Advances in technology, international power shifts, new forms of decentralized organization, and the long-term consequences of climate change are key drivers shaping the future battlefield. Therefore, future military action needs to put more emphasis on accelerating data gathering, analysis, and application, requires a comprehensive recognized operational picture, depends on shorter sensor to shooter cycles, and demands more flexible and partially automated measures of response. Strategic depth and delivering effects at greater distance become more important and should go hand in hand with developing counter-Anti Access/Area Denial (A2/AD) capabilities.

The Chief of Defense’s (CHOD) 2022 operational guidance (Generalinspekteur 2022) underpins these reflections. It calls for new innovative solutions to be “battle ready” and highlights the pressing need to assess sensor data “on the edge” to counter adversarial jamming. Emphasizing the need to transmit, store and make data available and usable, this document further argues that AI will become increasingly important to “support data correlation and data processing on behalf of the Commander and his command aides” (Generalinspekteur 2022: para 295, 297–98).

The CHOD’s guidance also embraces the idea of Multi-Domain Operations (MDO). Although not new as a core idea, the document argues, today’s focus provides opportunities to link capabilities across domains, advance operational tempo, and impose dilemmas on the adversary by precise direct and indirect effects (Generalinspekteur 2022: 286, 290). Therefore, the MoD’s Directorate-General for Planning has tasked the Planning Office of the Bundeswehr at the end of 2022 to start the national MDO implementation and submit the respective concept document to the CHOD by mid-2024 (Interview, 28 February 2023).

1.3 Digitalization and Software-Defined Defense

Today’s emphasis on digitizing armed forces originates from Network-Centric Warfare or Network-Enabled Operations, the state-of-the art military concept in the early 2000s. To realize the Bundeswehr’s unique selling proposition—preparing and using armed force—Network-Enabled Operations and the ability to contribute to MDO are key. That’s why a seamless and powerful ICT federation is indispensable (BMVg 2017) to make the armed force more assertive, increases the Bundeswehr’s operational capability as a whole and on the digitized battlefield, and support administrative action (Färber 2023: 225).

Currently, defense digitalization morphs into software-defined defense, a concept that detaches the hardware aspects of military capabilities from the software aspects with the goal to connect the latter in “data-centric, multi-modal, multi-domain, adaptative battle networks” (Soare et al. 2023: 2). Gen Michael Vetter, the MoD’s Chief Information Officer, backs software-defined defense as a new way of ensuring future capability growth by “digitally upgrading” legacy systems (Welchering 2023).

This understanding also underpins the MoD’s 2021 data strategy (BMVg 2021) and shapes its AI approach (Prenzel et al. 2023). As the contributions in this volume show, the Bundeswehr joins the chorus of many other armed forces in calling data “an asset of significant value” that enables information and effects superiority. Therefore, the data strategy is geared towards providing data of high quality and accessibility to strengthen the mission-readiness and resilience of IT and weapon systems, reduce life-cycle costs of these systems, boost the use of data across the Bundeswehr, increase the use of data, and enable data analytics.

This vision, however, is not yet in line with reality. Today, the Bundeswehr effectively operates in two worlds—old and new—requiring soldiers to envision the future while procrastinating future concepts and projects into legacy procedures and processes as the only means available to induce incremental change. This creates obvious tensions between “feel-good” digitalization, operated to convey the image of a techno-savvy and attractive employer, and defense digitalization meant to meet the Bundeswehr’s operational performance requirements (Interviews, 25 March 2022 and 6 February 2023).

1.4 Defense AI

1.4.1 Joint Thinking

Defense digitalization provides the umbrella for defense AI. The 2019 defense AI concept paper (BMVg 2019a) is Germany’s current capstone document, discussing in detail the general goals of using defense AI, the operationalization for the Bundeswehr, as well as the requirements (e.g., organization, human resources, legal aspects, IT hard/software aspects). In doing so, the paper follows the data-centric approach outlined above and defines

AI as a technology that uses machines with sophisticated algorithms taking on tasks that require – some sort of – intelligence to accomplish tasks that have previously required primarily or exclusively human decision-making or action (BMVg 2019a: 6).

The document envisages a holistic defense AI approach taking into account political, military, and industrial aspects (BMVg 2019a: 10–11). On industrial aspects, however, the MoD pushes itself to the backseat by arguing that “AI is not an explicit military capability, and the Bundeswehr is not the driver of AI-related innovation” (BMVg 2019a: 13). As this document puts civil and commercial developments ahead of defense AI, it is not surprising, that the document mainly refers to gains in efficiency, effectiveness, and process improvements as the key imaginaries describing the goals of using defense AI (BMVg 2019a: 17).

The document discusses potential areas of applications of defense AI but refrains from describing how exactly defense AI is expected to enhance the Bundeswehr’s key capabilities. Rather it argues that AI should be introduced with the help of broadly defined pilot projects that can quickly expose the Bundeswehr to defense AI (BMVg 2019a: 15), without specifying which capability areas should be targeted to achieve what kind of capability gain. Therefore, the strategic rationale underpinning defense AI is fuzzy, and it is unclear how defense AI will augment capability growth over the next decades. Consequently, there is a gap between high-level guidance and ongoing projects. The Armed Forces Digitalization Center is meant to close this gap with a new defense AI implementation strategy (Interview, 6 February 2023), which has not yet been published at the time of writing in January 2024.

1.4.2 Service Thinking

In this context defense AI ambitions of Germany’s military services diverge. In 2017, the Germany Army (Kommando Heer 2017a, b) published a series of concept notes outlining the future digital battlefield and the role of AI, followed by a fully-fledged defense AI position paper by the German Army Concepts and Capability Development Center (Amt für Heeresentwicklung 2019) two years later. It argued that that defense AI would help render basic services more efficient, improve combat-ready capabilities, and overcome existing capability gaps. In so doing the paper reflects US and NATO discourse on the digital and accelerated battlefield (hyperwar), talks about operating at machine speed, taking decisions on the edge, and using AI to coordinate and synchronize a growing number of sensors and effectors. Moreover, the paper also outlines how to further improve defense AI for human resources and material management and to enhance training and education (Amt für Heeresentwicklung 2019: 5–7, 12; Brendecke et al. 2020).

Without AI, today’s means of command and control will be insufficient to operate air power within the next 15 years, the German Luftwaffe contends. The Luftwaffe expects defense AI to synchronize information for Recognized Air Pictures, optimize flight routes, mission planning and mission management, coordinate target acquisition, and submit proposals on how to design and implement plans of attack. The service is taking baby steps in using defense AI, but its thinking is highly aligned with the US Air Force vision of using AI to set up Advanced Battle Management Systems to enable Joint All Domain Operations (Autorenteam Luftwaffe 2021).

So far, defense AI has played only a subordinate role for the German Navy. The service considers itself as technology driven as the Luftwaffe, but dire savings plans have limited the Navy’s capability development priorities to what is absolutely needed to ensure its survival. In the past, this even led the Navy to divest leading technology applications from aboard its ships to save costs. The current Navy leadership aspires a conceptual turnaround that provides more leeway to innovation, e.g., by establishing a new innovation cell with the Navy leadership team. As more defense AI use cases become known, sailors seem to become more aware how defense AI could offer added value aboard their ships (Interviews, 23 February and 14 March 2023).

The Cyber and Information Domain Service has an instrumental understanding of defense AI that is directly related to its core tasks. As this service plays a key role in providing common operational pictures (COP) it emphasizes the role of AI in providing analytical support and rendering digital processes more efficient and effective. As such, AI is part of the so-called “Analytics and Simulation” cluster which combines different methods such as pattern recognition, decision support, machine learning, and simulation (Bundeswehr 2020a; Färber 2023).

1.5 Ethics and Defense AI

Germany’s strategic culture implies that ethics plays a formative role for developing and using defense AI. This also explains the focus on arms control to shape the use of AI and other emerging technologies (SPD/Bündnis90/Die Grünen/FDP 2021: 146). In addition, the MoD’s 2019 defense AI concept unmistakably posits that the MoD needs to engage in a multi-stakeholder process to shape the public discourse on AI as—on its own—it can “neither lead nor shape the societal debate on AI as well as its risks and benefits, because the military use of AI constitutes only a small portion of a much broader topic” (BMVg 2019a: 10).

In this regard, leading officers opine that defense AI should always play a subordinate role to human decision-makers, who need to remain in control “of decision making as AI cannot replace human innovation, surprise, human values, personal experience, trust and emotions, and camaraderie” (Bock and Schmarsow 2023: 154). Ultimate human control is also paramount to prevent the rise of lethal autonomous weapon systems that Bundeswehr leaders reject (Ehlke 2021: 18). Furthermore, asked about the strategic purpose of using defense AI, one interview partner argued that it is all about “humanitarian precision,” a rhetorical figure that combines the reality of a post-heroic and risk averse society with the need for speed on the battlefield (Interview, 22 February 2023). As Jensen et al. (2022: 35–36, 40–46) have argued, such statements show that the Bundeswehr leadership holds a collective vision, which stabilizes institutional thinking about defense AI and narrates the socio-technical imaginary that shapes Germany’s dealing with defense AI. This also shapes the defense industry’s thinking, as the 2023 position paper of Germany’s leading, defense-relevant associations illustrates (BDSV et al. 2023).

But demanding respect for ethical principles and implementing them via technology development are two different things. On this very specific aspect the MoD’s guidance has so far remained vague, while in practice different initiatives emerge. At the international level the new ISO/IEC/IEEE 24787-700:2022 standard defines a process for value-based engineering (Hofstetter and Verbovszky 2023), which also informs the work of the NATO Data and Artificial Intelligence Review Board (DARB, Interview, 7 February 2023) and the German defense AI project GhostPlay (see below). At the national level the German Association for Electrical, Electronic, and Information Technologies (VDE) has submitted a cross-industry standard to ensure AI trustworthiness that shall lead to an AI Trust Label (VDE 2022). At the corporate level companies work on project specific solutions. One example is the FCAS Ethical AI Demonstrator envisaged to provide a scenario-based simulation environment to illustrate ethical dilemmas and possible solution options (Koch 2022).

2 Developing Defense AI

The German MoD and the Bundeswehr have kicked off numerous projects, but it remains difficult to understand how individual projects will contribute to future capability growth. In addition, structural pacifism has led to a bifurcated national ecosystem favoring knowledge stovepipes rather than an integrated approach.

2.1 Development Priorities and Projects

At the time of writing, a national defense AI capability roadmap has not been publicly released. Consequently, this section provides our assessment of more than a dozen ongoing projects that have been selected to illustrate the diversity of current activities.Footnote 1 We structure these projects along the Bundeswehr’s capability value chain—with some projects crossing several capability areas—and highlight the primary domain on which the respective projects focus:

  • Command, Control, Computers, Communications, and Cyber (C4/C5)

COPs are considered central to acting swiftly and precisely (Generalinspekteur 2022: para 271). Therefore, using AI is important especially with regard to assessing mass data, advancing pattern recognition, and computing suggestions for courses of action (BMVg 2019a: 17–18). This focus is also considered inconspicuous and in line with the dominant socio-technical imaginary thus giving the Bundeswehr freedom to explore AI’s strengths and shortfalls (Interview, 28 February 2023). Furthermore, the Bundeswehr’s military services have specific COP needs. The Navy, for example, wants to create a subsea situational picture by fusing data from various military sensors with geoinformation and information about key offshore and subsurface infrastructure. AI is meant to be used for object recognition, modelling, and new modes of data visualization (Interview, 23 February 2023). Space Situational Awareness satisfies a similar need for a different domain with BWI and the Cyber Innovation Hub of the Bundeswehr exploring the use of defense AI to forecast space weather and project orbital movements of objects to avoid space collisions (BWI 2021).

On a separate but related track the Luftwaffe’s AirC2 project evaluates the contribution of AI in increasing C2 efficiency and tempo and enhancing air C2 education and training. In addition, the Air Combat Management System project evaluates the use of AI to anticipate adversarial action, produce recognized air pictures, and recommend future courses of action (Interview, 25 March 2022). In view of future air power, defense AI is also a major issue for the Future Combat Air System (FCAS) and the Next Generation Weapon System (NGWS). FCAS has identified a total of eight use cases for defense AI. Situational awareness with AI shall “support orientation, decision making, and planning; either for a human operator using tactical displays or for automated functions directly assessing (…) digital data” (Azzano et al. 2020).

  • Intelligence, Surveillance, and Reconnaissance (ISR)

Project MITAFootnote 2 focuses on wide area surveillance with the help of an AI-augmented sensor grid and automated data fusion. The goal is to produce a COP that illustrates adversarial troop movements in 3D and identifies adversarial intruders in real-time (BWI 2022). AI for ISR is also of interest for the German Navy, which is developing AI-augmented solutions to assess sensor data and classify hydroacoustic signatures in cooperation with the University of the Bundeswehr/Hamburg (Written communication, 22 July 2022). Furthermore, BWI and the Navy have launched KALMAR in cooperation with the startup marinom to use AI to advance the Navy’s underwater situational awareness (Tedeski 2023).

  • Precision Effects

Based on GhostPlay (see below) Wild Hornets supports the German Army Concepts and Capability Development Center to develop tactics for air-launched effector swarms that target an adversarial high-value asset and test the feasibility of using air-launched effectors against next generation ground-based air defense solutions (Henckel 2023).

  • Support

Several FCAS use cases address AI for supporting functions, for example, to enable complex guidance and flight control behavior to navigate unmanned platforms, improve anomaly detection, systems operator training, and improve production, maintenance, and logistics, thus reducing life cycle costs (Azzano et al. 2020). The University of the Bundeswehr/Hamburg uses AI simulation and numerical modelling to advance existing test and validation methods to improve the electromagnetic resistance of unmanned systems. The MissionLab at the University of the Bundeswehr/Munich tests mission planning/management systems, intelligent sensor systems or adaptive assistance systems with experimental simulation and flight trials thereby also using AI.

  • Cross-Functional Projects

Advancing situational awareness and situational understanding by improving the C2-ISR link with defense AI is a key national R&T priority for NGWS with a focus on sensor data fusion, sensor resource management, and the integration of both elements. This project also explores options for a so-called AI Backbone that would provide a “single set of algorithms” to support different tasks and establish an open and unitary framework to facilitate the comprehensive use of defense AI (Interview, 2 March 2023). From 2019 to 2021 the program office for the German-Franco Main Ground Combat System (MGCS) ran a project with industry partners to use several unmanned aerial vehicles (UAV) as sensor carriers to produce a recognized operational picture integrated into the Army’s C2 system via SitaWare Frontline (Wiegold 2019; ESUT 2021). A comparable project looks at the role of AI in automatically assessing and incorporating terrain specifics into operational planning with the aim of using terrain features for tactical advantages (Interview, 14 November 2022). In addition, URANOS KI, a follow-on project to MITA, combines defense AI for modular effector systems with a UAV-based surveillance system. Automating the handover of targeting data to different effectors is one of the capabilities to be developed (Interview, 14 November 2022; Prenzel et al. 2023: 43). Finally, project GhostPlay develops defense decision algorithms (Play) for defender and aggressor tactics thereby using a powerful simulation environment (Ghost). GhostPlay started with a Suppression of Enemy Air Defense (SEAD) scenario using UAV swarms to target a high value asset protected by ground-based air defense. Attacker and defender use AI developed tactics to outsmart each other. The “ability to learn tactical behavior in cooperation with other machines and/or humans” constitutes the project’s AI research focus (Borchert et al. 2022a).

2.2 Germany’s Defense AI Ecosystem

The German government (Bundesregierung 2019: 3) considers AI a defense-relevant national key technology, but it is unclear, what this categorization implies. This is relevant because Germany’s bifurcated techno-industrial ecosystem mostly segregates defense-relevant actors from the rest (Borchert et al. 2022b; Hagebölling and Barker 2022: 6). Therefore, the German Bundeswehr has only access to a limited spectrum of the country’s techno-economic power. Overall, the defense AI ecosystem rests on four building blocks:

  • Bundeswehr

The Bundeswehr has established new entities to advance defense digitalization, such as the so-called Systems Centers (Systemzentren) for single services. Some service-specific institutions also have a cross-functional task, such as the Center for Digitalization of the Bundeswehr. This center is key to develop Germany’s CIR capabilities, provides software development and IT integration capabilities for the Bundeswehr, and oversees developing the Bundeswehr’s capabilities for military intelligence, electronic warfare, and geoinformation (Bundeswehr Undated). Other institutions help spinning-in digital solutions from outside the Bundeswehr (Cyber Innovation Hub) and supporting digitalization by maintaining key infrastructure and application development (BWI). Finally, the German Army also uses its test and experimentation unit as a testbed for rapid technology insertion and experimentation as well as synchronized concept and technology development (Bundeswehr 2020b).

  • Research and Technology Organizations (RTO)

RTO constitute the second pillar of the defense AI ecosystem. Here bifurcation becomes most obvious. More than 70 universities and universities of applied sciences adhere the voluntary civil clause that prevents them “from engaging in defense research and cooperating with the defense industry” (Borchert et al. 2022b: 437). This also means that the Bundeswehr will not directly benefit from the German government’s decision to set up six centers of competence on AI and fund “the establishment of 100 new professorships in AI at German universities” (Federal Government 2020: 10). Furthermore, the German Research Center for AI (DFKI), which has been pioneering AI research since the late 1980s, does not engage in defense either. To some extent the Bundeswehr can close the gap by relying on R&T conducted at its universities in Hamburg and Munich. Activities at these two locations have received a boost thanks to a €500M budget to set up the Digitalization and Technology Research Center (dtec.bw) that is meant to advance defense digitalization.Footnote 3 Beyond dtec.bw, the lion’s share of Germany’s defense research falls on the Fraunhofer Society and the German Aerospace Center.

  • Old and New Defense Industrial Players

The defense industry forms the third pillar of the defense AI ecosystem. Most of Germany’s well-established defense companies like Airbus, Atlas Elektronik, Hensoldt, KNDS or Rheinmetall are involved in developing or adopting AI for defense purposes in one way or another. More recently, several new players and startups (BMWK 2022) with a dedicated focus on AI and defense AI have entered the market. Some of them originate from the commercial world and join forces with incumbent defense players:

  • 21strategies specializes in developing large scale multi-agent reinforcement learning to compute optimal decision-making strategies under uncertainty in the context of national security, capital markets, and supply chains. 21strategies works on GhostPlay, Wild Hornets, FCAS, and NGWS. Hensoldt cooperates with 21strategies.

  • Aleph Alpha is working on large language models and develops generative AI solutions to support public and private sector applications. Among others, Aleph Alpha is working on defense AI solutions for FCAS.

  • Data Machine Intelligence Solutions develops data modeling and visualization solutions, inter alia, with a focus on solutions for mission planning and management as well as simulation technologies. Data Machine Intelligence Solutions also contributes to FCAS’s defense AI work stream.

  • HAT.tec focuses on developing technologies in support of human-autonomy teaming, with a focus on automated reasoning, planning and decision-making. HAT.tec also works on defense AI solutions for FCAS.

  • Helsing develops AI for real-time information processing and turning unstructured sensor data into common operational pictures. The company is headquartered in Germany with subsidiaries in France and the United Kingdom. Helsing works on defense AI for FCAS, NGWS and MITA. Helsing cooperates with Rheinmetall Defense Electronics, Saab, and MBDA.

  • Traversals uses AI for open-source intelligence to analyze and assess global events, identify potential threats, and assessing multilingual information. Traversals AI Dynamic Frontline Monitoring, for example, uses AI-enhanced technologies to provide a 24/7 near-real time operational picture of the Ukrainian-Russian front line.

  • IT and Consulting Companies

IT and consulting companies such as Accenture, Atos or SAP form the final pillar of the German defense AI ecosystems. These companies are instrumental in supporting concept development, providing hardware infrastructure and computer processing capacities as well as assisting the synchronization of digitalization and organizational change.

3 Organizing Defense AI

The Bundeswehr is in its early days to adjust its organizational fitness to future defense AI requirements. The 2019 capstone document acknowledges that a strong Bundeswehr-common approach with joint responsibility for capability development is needed to counter the risks of duplication, parallel structures, crowding-out effects, and fragmentation (BMVg 2019a: 20). So far, however, tensions exist between top down-driven and decentralized service-specific approaches.

3.1 Joint Approaches

Defense AI is part of the digitalization agenda set by the Directorate-General for Cyber/IT (CIT). In 2019 the German MoD has also established a Digital Council (Digitalrat), which advises the Minister of Defense and provides impulses to advance defense digitalization (BMVg 2019b: 16). Additionally, the Directorate-General for Planning implements the Bundeswehr’s integrated planning. Since defense AI is part of the toolbox needed for the Bundeswehr’s future development, this Directorate-General provides the leading desk officer for defense AI. He also chairs the Bundeswehr’s defense AI community, a semi-formalized network for information sharing.

Tensions arise from the fact the Bundeswehr’s military services follow different digital levels of ambition and enjoy great leeway in implementing their respective digital agendas while CIT shapes the broad guidelines and the idea of a Bundeswehr-common AI backbone. This creates a “wait and see” atmosphere as the services need to strike a balance between following through on their own agendas and supporting a joint agenda, which might come at the cost of sacrificing service-specific resources for joint tasks. A cluster approach respecting joint and service-specific interests could work but very much depends on the willingness of the actors involved and the availability of extra resources (Interviews, 25 March 2022 and 14 March 2023).

3.2 Single Service Approaches

Against this background, the Army is digitalizing land-based operations (D-LBO) to create a whole-of-service digital federation for future operations. Its 2019 AI concept paper envisions setting up an AI steering group with the Army Command to oversee the work of the so-called Army AI Work Bench at the Army Concepts and Capabilities Development Center. This work bench would serve as the overall coordination mechanism for all Army AI activities and liaise with industry and academia. In addition, the Army would create a development center mainly focusing on training defense algorithms and developing key data models as well as an AI data center that would take care of Army data, provide data expertise and data scientists (Amt für Heeresentwicklung 2019: 14–15/19–20; Dani 2022). Elements of this vision will be realized with the Army’s forthcoming Systems Center for Digitalization. It is likely to be the powerhouse for all things digital of the Army and play an important role by strengthening sovereign defense software development (Interview, 22 February 2023).

The German Luftwaffe is exploring the impact of defense AI on future air power. So far, the service has taken organizational baby steps with one desk officer in the Luftwaffe Command overseeing the subject matter. The Luftwaffe also considers defense AI as part of its broader digitalization agenda and as an important enabler to advance air power innovation. Tensions exist as the service has two responsible officers for digitalization (Deputy Air Chief) and innovation (desk officer, LTC level). Both have pledged to inform each other but given “split” responsibilities true leadership on defense AI remains yet to be developed (Interview, 25 March 2022).

The new Chief of the Navy puts great emphasis on naval innovation. He has created the position of a Commissioner for Innovation, Digitalization, Empowerment, and Agility (ID:EA, Marinekommando 2022) at the Naval Command. This new position is to bridge the digitalization/innovation divide and push both agendas. Defense AI is part of the ID:EA tasks and will benefit from a vast network of naval reservists that is to be expanded. Overall, the current focus is on breaking up existing structures by creating opportunities for new digital naval projects outside existing planning processes that are considered too cumbersome to deal with (Interviews, 23 February and 14 March 2023).

Finally, the Cyber and Information Domain Service operates and protects the Bundeswehr’s IT infrastructure, is engaged in electronic warfare, provides satellite-based imagery reconnaissance data, and operates the Bundeswehr Geoinformation Center. Its Center for Bundeswehr Digitalization and Cyber and Information Service Capability Development pools software analysis and software development expertise. Regarding defense AI, the Electronic Warfare Battalion 912, for example, plays an important role as its own AI laboratory is exploring the use of AI to calculate flight paths or analyze radio communications (Fleischmann 2022).

4 Funding Defense AI

Whereas the German government has pledged to spend €5bn until 2025 to implement the national AI strategy, it is difficult to gauge how much the Bundeswehr spends on defense AI. Overall, investments in Germany’s digital defense infrastructure were set to rise with roughly 20% of the €100bn Sondervermögen (special fund) originally to be spent on this priority. Among other projects, this included around €8.5bn for the Army’s landmark D-LBO or €2.6bn for the German Mission Network. However, changes in the plans for the special fund make it difficult to assess whether this is still true. The 2023 budget law cut defense R&T by €200M to €330M whereas spending on defense development and experimentation stood at around €515M. Fortunately, the 2024 draft budget significantly increases R&T spending again to €565M in the regular budget, with an additional €50M coming from the special fund (Deutscher Bundestag 2023: 2199/2226). However, development and experimentation spending decrease to €215M in 2024. Furthermore, the MoD can spend around €40M (2023) and €50M (2024) respectively on methods such as Concept Development & Experimentation, modeling and simulation and innovation competitions, and around €25M on disruptive innovation in cybersecurity and key technologies (Bundesgesetzblatt 2022a: 1034; Bundesgesetzblatt 2022b: 45/48–49/69–70; Deutscher Bundestag 2023: 2201).

No public figures are available for spending on defense AI. The law on the Sondervermögen originally earmarked a total of €422M for research, development, and AI, with AI focusing on surveying and safeguarding wide areas (Eastern Flank) (Bundesgesetzblatt 2022a). The 2023 Sondervermögen budget law breaks this focus area down to a €16M increment without further specifying the AI amount. This budget line significantly increased in 2024 to €667M, but no further breakdown is publicly available. This suggests that a large part of these funds will most likely be used to partially offset the decrease of the experimentation budget in the regular defense budget (Deutscher Bundestag 2023: 2227). In addition, several dtec.bw projects focus on developing defense AI. Taken together, the total four-year budget of three AI projects at the University of the Bundeswehr/Hamburg is about €20–30M, or €5–7.5M per year. The MissionLab at the University of the Bundeswehr/Munich operates on a total budget worth around €20M, or around €5M per year. In addition, we speculate that the German MoD spends a lower two-digit million amount per year on developing AI for NGWS. Considering these figures and adding a calculated reserve for projects unknown at the time of writing, we assume that the MoD currently spends around €50M per year on defense AI software development.

5 Fielding and Operating Defense AI

In-service defense systems use AI applications but a clear delineation between software-enabled analytics and automation and proper AI is difficult to draw. Consequently, the true status of the Bundeswehr’s fielding and operating of defense AI remains opaque. Overall, the following overview of selected projects is in line with the development priorities discussed above:

  • Command, Control, Computers, Communications, and Cyber (C4/C5)

The German MoD’s Directorate-General for Strategy and Operations uses the so-called Preview system to analyze open-source intelligence (OSINT) for early warning with AI for data analytics and predictive analysis. Users can zoom in on each of the more than 60 indicators’ quantitative assessment and track assessment changes over time. The system also provides access to the sources that underpin assessment results. Preview is a multi-language solution that also offers back casts to validate the feasibility of current assessments with a database reaching back to 2015. As Preview is an OSINT solution, classified and open date are not yet fused. In addition, a link between crisis early warning and the Bundeswehr’s activities on establishing COPs has yet to be established (Interviews, 18 and 21 March 2023).

  • Intelligence, Surveillance, and Reconnaissance (ISR)

In early 2022, the Bundeswehr decided to equip all NH90 helicopters with Hensoldt’s Kalaetron Radar Warning Receiver, which uses AI for big data analysis to quickly detect new threat patterns and with a very low false alarm rate (hartpunkt.de 2019; EDR Magazin 2022). Since 2018, BWI has been experimenting with the use of AI in combination with radar technology already in use on construction sites to see through walls to detect humans as individuals or groups and identify the current state of motion (Ilg 2022).

  • Precision Effects

The Luftwaffe uses Diehl’s IRIS-T air-to-air missiles equipped with intelligent image processing to detect and ignore adversarial infrared decoys when engaging a target (Penney 2000). Similar technologies likely underpin the image-scanning seeker of the Rolling Airframe Missile (RAM) that the German Navy is using to protect frigates and class K130 corvettes (ESUT 2022b; Naval Technology 2014). In addition, Saab’s Arexis electronic warfare sensor suite including AI algorithms by Helsing has been selected to upgrade Eurofighter jets (Saab 2023).

  • Support

The Bundeswehr’s Joint Support Service has been experimenting with the use of AI for an early warning system to support national crisis management and to support warehouse functions (Bundeswehr 2021; ESUT 2022a). The Medical Command uses civilian AI applications for decision support of doctors in the fields of analytics, diagnostics, and individual therapies (BMVg 2019a: 11). In addition, BWI has been developing BundesWEAR, an app with AI features that offers individual measurements, suggests the best fitting clothing size, and offers online orders for home or barracks deliveries (ESUT 2022a).

6 Training for Defense AI

The Bundeswehr is in the very early stage of exploring the impact of defense AI on training. The MoD’s 2019 defense AI capstone document argues that future members of the Bundeswehr will need broad and specialized AI expertise, software development know-how, improved MINTFootnote 4 knowledge, and interdisciplinary know-how to develop solutions for human-machine interaction (BMVg 2019a: 21–22). At the joint level, the Bundeswehr Command and Staff College (Führungsakademie) is responding. In view of launching the new curriculum in autumn 2024, stocktaking is underway to define the future role defense AI is going to play—both as an instrument for training and education and as a subject that future officers need to understand. In April 2023, the College launched a digital open space learning environment. The modular set up could also be used to create interfaces for integrating wargaming, serious gaming, and AI-enhanced training solutions.

In parallel, the University of the Bundeswehr/Hamburg is working on a new AI bachelor’s and master’s degree course. The program aims at teaching technical basics in the fields of mathematics and informatics as well as adjacent technology areas such as sensors, acoustics, or information technology. The new program would also embed defense-relevant AI in the broader societal context with building blocks focusing on law, ethics, sociology, and political science (Interview, 25 February 2023).

Complementary service-level activities are under way. The Army is looking at the role of defense AI in live and constructive training simulations as well as AI-augmented learning analytics to adjust teaching to individual learning progress (Interview, 7 March 2023; Amt für Heeresentwicklung 2019: 12). The Luftwaffe is exploring AI for new tactics, techniques, and procedures related to air defense and dogfight scenarios and looks at using AI to train Luftwaffe operators in advancing and improving planning cycles as part of its AirC2 project (Interviews, 7 and 25 March 2022). Not yet using defense AI for training, the Navy is mulling the idea of using AI-based training for sea-based signals intelligence (Interviews, 22 February and 14 March 2023).

Additionally, different initiatives have been launched to train defense AI algorithms. The Luftwaffe has procured an off-the-shelf software product to teach air power gaming. While the primary purpose was to improve the respective planning and operating procedures, data generated by using the software is also used as a basis to train future defense AI algorithms (Interview, 25 March 2022). The Army is working on using simulation-based training with reinforcement learning to train neural networks with the goal of enhancing the autonomous behavior of unmanned systems in battlefield scenarios. It also looks at reinforcement learning to train neural networks to command battlefield units. Another focus area of the Army emerges from the need to generate training data for AI-enhanced image recognition (Interview, 7 March 2023).

7 Conclusion

Today, German defense AI is a grassroots movement. Motivated people push projects to bring defense AI into the Bundeswehr. Structural and procedural provisions are in place to enable change. Overall, however, change is first and foremost about closing fundamental capability gaps. The Bundeswehr may want to operate at the technological edge, but existing shortfalls inhibit Germany’s armed forces from doing so. New concepts that leverage emerging technologies are bound back by the resistance of a Bundeswehr bureaucracy solidly grounded in the status quo.

This is no surprise. As we have argued, Germany’s defense AI approach is locked in a “master and servant” logic deeply rooted in the country’s strategic culture and organizational set up (structural pacifism). Consequently, Germany prioritizes security and technology policy options, which comply with its non-belligerent post-war identity. Domestic socio-political legitimization of the use of force is consistently more important than the impact that can be achieved by using it.

This preference will undoubtedly continue to determine political visions on future roles of defense AI and the panorama of technologies deemed acceptable for military use. This narrows future development options and limits the impact of defense AI to an evolutionary trajectory from the start. Since innovation in a tightly regulated defense market relies heavily on capability-pull by the Bundeswehr, its muted technology appetite does not bode well for the defense industry, too.

Consequently, broadening the footprint and strengthening the influence of the defense AI grassroot movement will require the MoD to do some heavy lifting along four lines of effort. First, the Bundeswehr needs to be more precise on how defense AI will boost its capabilities. This guidance should delineate Bundeswehr-common and service-specific defense AI capability goals and identify current defense AI shortfalls. Prioritizing mitigation measures can lead to creating a roadmap to address these shortfalls via national or multinational R&T projects and procurement programs.

Second, a defense industrial policy for AI is needed. By arguing that it does not drive technological development, the Bundeswehr effectively drops out as a demanding launch customer for cutting-edge defense AI solutions made in Germany. This creates distorted market signals for innovative companies that might consider entering the defense business. It is therefore high time for the MoD and industry to specify the “terms and conditions” on which the future defense data ecosystem will operate. This requires striking a balance between the Bundeswehr’s interest in unrestricted data access, industrial preferences for data monetization, and the general need to incentivize a data sharing dynamic that also involves stakeholders that have not participated in generating original data (Datenethikkommission 2019: 145–148).

Third, Germany needs to define its international defense AI ambition, for example, by positioning itself as a defense AI framework nation. A hardware-focused framework nation could advance multinational defense AI by offering international partners access to its new digital battlefield infrastructure. Alternatively, a software-focused framework nation could zoom in on specific applications, for example, by offering AI-enhanced red teaming to support multinational capability development and design evaluation for multinational projects like FCAS and MGCS. Moreover, the Bundeswehr could turn its strong focus on ethics into an asset by combining value-based engineering with simulation to offer partners a new digital test lab for the responsible use of defense AI.

Finally, the Bundeswehr needs to consider how to certify, qualify, and admit future defense AI solutions. This is a daunting task because today’s system benefits the original equipment manufacturers (OEM) in their role as gatekeepers that can resist modifications of existing defense products (Interview, 14 March 2022). This is likely to undermine software-defined defense if software-induced modifications change the overall characteristics of a defense solution for which the OEM—not the software developer—bears ultimate responsibility. Although there is no easy way out of this dilemma, an AI-enhanced simulation environment could provide an option to test the characteristics and the performance of future defense AI solutions. The resulting digital twin could be augmented, for example, with mission-critical data gathered during international Bundeswehr operations as well as AI-enhanced red force elements.