3 of 5 - The Antifragile War: The Iranian Mosaic Doctrine, Agentic AI, and How to Prevent a Perpetual War

Part 3 of 5: A New Frame of Reference: Policy Levers for an Antifragile War

"If you could see your whole life from start to finish, would you change things?" - Arrival

The first two parts of this series argued that Iran’s mosaic doctrine is designed to turn pressure into adaptation, and that layering agentic AI, autonomous targeting, and automated production onto that mosaic compresses time and lowers the cost of sustained coercion [1][3]. The practical question is what follows; if decapitation logic misfires and acceleration risks a self‑sustaining conflict loop, what kind of policy playbook can advanced democracies use to lengthen substitution timelines, raise the cost of regeneration, and slow the machine without inflaming the system?

Here we treat the mosaic not as a puzzle of individual targets, but as a living system whose resilience depends on three families of enablers (software and design diffusion, shadow finance and logistics, and decentralized doctrine and verification). The core claim is straightforward; policy that focuses on outputs (launchers, depots, even named commanders) without systematically constraining the enablers will struggle to change the trajectory of the war. By contrast, sustained, multilateral pressure on enablers can lengthen substitution cycles, dampen the antifragile feedback loops, and create space for the diplomatic work that Parts IV and V will explore.

From Targets to Timelines

Parts I and II describe a system that expects to lose nodes and has already built social, financial, and doctrinal pathways to replace them [1]. Recent U.S. and allied sanctions illustrate this dynamic. Monetary policy measures and diplomatic engagements have repeatedly designated procurement networks supporting Iran’s missile and UAV programs across Asia, the Gulf, and Europe [4][5], only to see reconstituted networks emerge through new front companies and shipping routes. This is an analytical inference grounded in public sanctions patterns: each enforcement wave clears part of the network graph but does not fundamentally change the time it takes for the mosaic to restore capacity.

In an antifragile architecture, the relevant variable is not “Can we destroy X?” but “How long does it take for the system to substitute for X, and what does that substitution teach it?” When sanctions push procurement into more sophisticated channels, and when strikes prompt decentralization of production or deeper embedding in civilian infrastructure, the system’s learning curve steepens. Without specific attention to substitution timelines, pressure risks optimizing the mosaic for long‑term survival. Policy design therefore needs a different objective function. Instead of maximizing the number of destroyed nodes, it should seek to:

1. Stretch the time between loss and substitution.
2. Reduce the quality of substituted capability.
3. Channel adaptation into less escalatory and more negotiable directions.

Those goals are harder to measure, but they align with how the mosaic actually works. They also map cleanly onto the three enabler families that sustain the antifragility of the Mosaic Doctrine.

Enabler One: Software, Design, and Technical Diffusion

Iran’s missile and UAV programs illustrate how software and design diffusion act as force multipliers for a distributed system. Open‑source reporting and government actions document how Iranian entities have exported UAV designs, licensed production, and shared technical know‑how with partners, enabling local assembly in Russia and elsewhere [2][9]. In this model, “production” is no longer a single facility; it is a network of designs, firmware, and modular components that can be recombined across jurisdictions.

Agentic AI and automation amplify this dynamic. As described in Part II, AI‑assisted guidance, swarm coordination, and automated production planning can increase the effectiveness of low‑cost platforms and optimize distributed manufacture [3]. This is an analytical inference built on emerging uses of AI in targeting and production. Where once design packages and software toolchains are mobile, shutting down a single plant matters less than constraining the circulation and upgrading of the underlying code and processes.

Policy has begun to move in that direction, but unevenly. U.S. export‑control moves and related commentary highlight that Dept. of Commerce has added Iran‑linked UAV and missile entities to its Entity List and expanded foreign‑direct‑product rules that touch Iran [9], signaling a shift toward software and design related controls rather than purely hardware embargoes. EU and allied measures targeting ports and logistics tied to missile and drone transfers likewise attempt to disrupt the broader ecosystem that moves controlled technologies [5][6].

Three principles follow for a more systematic approach to this enabler:

1. Treat design chains as strategic terrain: Export controls and sanctions should prioritize CAD files, firmware, AI toolkits, and modular production blueprints that enable rapid replication, not only the finished airframes or engines.
2. Align dual‑use governance with mosaic realities: Many AI and automation tools relevant to drone production and targeting are dual‑use; governance must focus on high‑risk capabilities and require safeguards from major providers [10].
3. Use positive incentives to channel diffusion: Attempts to block all military‑relevant AI are unlikely to succeed; shaping diffusion through governed pathways is more realistic.

These measures do not eliminate Iran’s ability to innovate, but they can slow the rate at which new software‑enabled tactics and designs diffuse through the mosaic, lengthening the substitution and learning cycles that underpin antifragility.

Enabler Two: Shadow Finance, Logistics, and the “Shadow Fleet”

Part I emphasized that the mosaic’s sustainment problem is fundamentally transnational; logistics, finance, and procurement cross borders and exploit uneven enforcement [1]. Recent sanctions underscore this point. US Treasury, Office of Foreign Assets Control (OFAC) has targeted a multinational set of entities; from aviation‑parts suppliers tied to the Iranian Aircraft Manufacturing Industrial Company (HESA), to firms underpinning oil‑smuggling “shadow fleets,” to networks serving missile‑propellant suppliers and UAV engine producers [4][5][6]. The U.S. State Department’s 2026 measures to disrupt Iran’s weapons procurement networks and shipping highlight how petroleum revenues and maritime logistics feed into missile and UAV programs [5].

Despite these actions, sanctions‑compliance and maritime‑analytics reporting shows that new shells, flags, and routes continually emerge [6]. Ships re‑flag, ownership structures shift, and trade intermediaries migrate to jurisdictions with weaker enforcement. Enforcement that does occur has tended to be episodic and node‑focused, while the networks evolve continuously.

A policy playbook oriented toward substitution timelines would emphasize:

1. Persistent, AI‑assisted financial enforcement: Financial‑intelligence units are beginning to use machine learning to map beneficial ownership and detect anomalous shipping behavior [6].
2. Harmonized maritime risk regimes: Coordinated AIS‑integrity checks, shared high‑risk vessel lists, and conditional port access can constrain maneuvering room for sanctioned fleets [6].
3. Ecosystem‑level sanctions and compliance incentives: Targeting clusters rather than single entities, paired with compliance support for private‑sector actors, reduces inadvertent facilitation [4][6].

Here again, the aim is not total interdiction but longer, more uncertain replenishment timelines. If it takes months, rather than weeks, for a procurement network to reconstitute (and if that reconstitution carries higher financial and legal risk) the mosaic’s ability to turn external shocks into rapid substitution is weakened.

Enabler Three: Doctrine, Delegation, and Verification

The third enabler is both doctrinal, cultural, and social. Iran’s mosaic doctrine embeds delegated authorities, pre‑planned succession ladders, and Basij‑based social integration to ensure that provincial IRGC commands and proxies can fight on even if central leadership is degraded [7][8][13]. Reporting on the current conflict shows how decentralized provincial commands, external proxies, and “mosaic defense” concepts are being used to preserve operational continuity after leadership losses [7][8].

AI-enabled technical acceleration across domains sharpens this problem. If AI‑assisted targeting and agentic decision‑support tools are pushed down to lower‑level commanders or proxies with limited political accountability, the risk of unauthorized escalation rises [3][14]. In such an environment, human‑in‑the‑loop governance is not only an ethical requirement; it is a strategic tool for re‑centralizing some measure of control over lethal decisions and the overall kill chain [10][12].

Three lines of policy effort stand out:

1. Norms and law around meaningful human control: Multilateral processes at the UN and elsewhere have begun articulating principles for maintaining human oversight of AI‑enabled weapons [10].
2. Verification architectures suited to mosaics: Traditional arms‑control verification is poorly suited to decentralized systems; mosaics require hybrid technical‑and‑local monitoring mechanisms.
3. Narrative and incentive design: Local communities embedded in Basij structures respond to incentives; linking de‑escalation to tangible benefits can shift behavior at the node level.

These doctrinal and verification measures prepare the ground for the diplomatic approaches to fragmented authority that the next article in this series will explore in-depth.

Multilateralism as a Systems Requirement

Across these three primary enablers, one pattern recurs: unilateral action struggles against a distributed, transnational system. Procurement networks span Iran, Türkiye, the UAE, China, Hong Kong, and Europe [4][5][6]. The maritime “shadow fleet” relies on flags and ports far from the Gulf [6]. The software and AI tools that enable acceleration are built and hosted by global firms [10]. For this reason, multilateralism is not a moral preference or a “nice to have”; it is a global conflict management systems constraint in the Age of AI. A mosaic that can route around any single jurisdiction will only feel meaningful pressure when multiple jurisdictions move in concert.

Human‑in‑the‑Loop as Strategic Brake

Human‑in‑the‑loop governance is both an ethical and strategic imperative [3], the strategic dimension emphasized here. When decision cycles compress to seconds, as U.S. commanders have described in the current conflict [14], the space for political judgement shrinks. In a mosaic where lower‑tier actors are pre‑authorized to act and where narratives of existential resistance are strong, machine‑speed operations magnify the risk of miscalculation and uncontrolled escalation. Requiring meaningful human control over targeting and escalation decisions is therefore a form of deliberate friction; designed to slow the machine enough for leaders to integrate legal advice, strategic assessment, and diplomatic signals before irreversible actions [10][12].

Seeing a New Picture

Part I diagnosed Iran’s mosaic doctrine as an antifragile military‑political system that converts pressure into adaptive capacity [1]. Part II showed how agentic AI, autonomous targeting, and automated production accelerate that system [3]. Here, we’ve ironed out that strategic success depends less on destroying visible outputs than on constraining the enablers (software and design diffusion, shadow finance and logistics, and decentralized doctrine and verification) that make rapid substitution possible.

Next, we will examine what negotiation and de‑escalation look like when authority is fragmented, mapping practical diplomacy with decentralized actors who sit inside the mosaic rather than above it. The core through‑line of this series remains constant: in an antifragile war/conflict, the most strategic moves are those that change how the system learns, substitutes, and decides; not just how it bleeds.

References

1. Parts I–II of The Antifragile War, March 20–21, 2026.
2. Jessica Seltzer and John Caves, “Alabuga Drone Plant,” Iran Watch, Nov. 8, 2024.
3. Robert Booth & Dan Milmo, “Iran war heralds era of AI‑powered bombing,” The Guardian, Mar. 3, 2026.
4. U.S. Department of the Treasury, “Treasury Disrupts Iran’s Transnational Missile and UAV Procurement Networks,” Nov. 11, 2025.
5. U.S. Department of State, “Sanctions to Disrupt Iran’s Weapons Procurement Networks and Shadow Fleet,” Feb. 25, 2026.
6. Windward, “OFAC Targets Iran’s Shadow Fleet and Weapons Networks,” Feb. 25, 2026.
7. Frud Bezhan, “With Top Brass Dead, Iran Deploys Decentralized ‘Mosaic’ Strategy,” RFE/RL, Mar. 7, 2026.
8. Al Jazeera, “The ‘Fourth Successor’: Iran’s plan for a long war,” Mar. 10, 2026.
9. Kharon, “BIS Targets Iran Drone Networks,” Oct. 8, 2025.
10. UN Regional Information Centre, “AI and the Dangers of Lethal Autonomous Weapons Systems,” Jan. 6, 2025.
11. Kian Meshkat, LinkedIn summary of OFAC/BIS actions, Apr. 17, 2024.
12. SIPRI, Autonomous Weapon Systems and AI‑enabled Decision Support Systems, June 2025.
13. Modern Diplomacy, “War Without a Center: Iran’s Mosaic Defense,” Mar. 10, 2026.
14. Dan De Luce et al., “U.S. military is using AI to help plan Iran air attacks,” NBC News, Mar. 11, 2026.