AI on the Battlefield: Project Maven, Drone Swarms & JADC2

Field notes. For engineers who want to understand why AI is more than a chatbot — it’s already a weapon, a sentinel, and the nervous system of modern warfare.

War Has Changed

In 2018, more than 3,000 Google engineers signed a protest letter and resigned. The reason: their company was building AI for military drone image analysis — Project Maven.

The letter read: “We believe Google should not be in the business of war.”

Three years later, the contract moved to Palantir. Project Maven continued — larger, more capable.

This isn’t a story about Silicon Valley vs the Pentagon. It’s a story about the reality that AI has entered the battlefield, whether we like it or not.

Project Maven — AI Eyes for Drones

Project Maven (formally the Algorithmic Warfare Cross-Functional Team) was launched by the US Department of Defense in 2017. Original objective: use AI to process thousands of hours of drone video from Middle Eastern operations.

Before Maven, video analysis required human analysts — slow, fatiguing, and unscalable as drone fleets grew. Maven brought computer vision to:

• Identify objects: vehicles, buildings, personnel
• Track movement across frames
• Prioritize which clips analysts needed to review first

flowchart LR
  I["Input<br/>Raw drone video feed<br/>thousands of hours/day"]:::input
  M["Maven AI<br/>Object detection<br/>movement tracking"]:::process
  O["Output<br/>Tagged clips<br/>geolocation metadata"]:::output
  A["Analyst<br/>Reviews high-priority targets"]:::human
  I --> M --> O --> A
  classDef input fill:#1e1b4b,stroke:#818cf8,color:#e0e7ff
  classDef process fill:#0f2a1c,stroke:#4ade80,color:#bbf7d0
  classDef output fill:#0f2436,stroke:#38bdf8,color:#bae6fd
  classDef human fill:#2e2410,stroke:#fbbf24,color:#fde68a

After the Google controversy, Palantir took over in 2019 via the Palantir Gotham platform — now widely used by the US Army for intelligence and logistics.

Fact: Palantir Gotham is used across 300+ US military and intelligence operations. In 2023, the US Army signed a new $250M contract for AI-powered logistics and targeting support.

JADC2 — The AI Brain for Multi-Domain Warfare

JADC2 (Joint All-Domain Command and Control) is the most ambitious concept in modern military history. Imagine this:

Aircraft, warships, ground forces, cyber units, and satellite weapon systems — all speaking the same AI “language,” in real time.

flowchart LR
  F35[F-35 air]:::asset
  DES[Destroyer sea]:::asset
  TNK[Tank land]:::asset
  CYB[Cyber ops]:::asset
  SPC[Space assets]:::asset
  JADC[JADC2 AI Layer]:::brain
  CMD[Commander Decision Support]:::cmd
  F35 --> JADC
  DES --> JADC
  TNK --> JADC
  CYB --> JADC
  SPC --> JADC
  JADC --> CMD
  classDef asset fill:#1e1b4b,stroke:#818cf8,color:#e0e7ff
  classDef brain fill:#2d1518,stroke:#f87171,color:#fecaca
  classDef cmd fill:#0f2a1c,stroke:#4ade80,color:#bbf7d0

The old problem: each military branch ran its own communication systems. In warfare, slow information = lives lost.

JADC2 aims to solve this with:

• AI-powered data fusion — unify sensors across all domains
• Predictive analytics — anticipate enemy movement based on patterns
• Decision support — surface recommendations to commanders in seconds, not minutes

This is no small project. US Congress has allocated over $1 billion for JADC2 since 2020, with key programs including the Advanced Battle Management System (ABMS) for the Air Force and Project Overmatch for the Navy.

DARPA ACE — AI Beats Human Pilot 5-0

In August 2020, in a quietly held test in Nevada, something remarkable happened.

DARPA ACE (Air Combat Evolution) — an AI trained for aerial dogfighting — faced off against an experienced F-16 pilot in an air-to-air simulation. Result: AI won 5-0.

The pilot described the AI as “aggressive” and “inhuman” — it had no fear instinct, no fatigue, and could execute maneuvers that would render a human pilot unconscious from G-force.

“It flies nothing like a human… it does things that, if I were to do them, I would have been unconscious.” — US Air Force pilot post-ACE test

In 2023, DARPA continued with the X-62A VISTA — a modified F-16 that flew autonomously in actual combat scenarios (without weapon systems, but in real airspace).

For engineers: Autonomous systems in safety-critical environments are no longer theoretical. They’ve been tested in real-world conditions.

Drone Swarms — A Thousand Eyes in the Sky

China: World Record, 1,000+ Simultaneous Drones

China Aerospace Science and Technology Corporation (CASC) demonstrated drone swarms exceeding 1,000 units in 2021 and 2023. These aren’t ordinary drones — they communicate with each other, distribute tasks autonomously, and can overwhelm conventional point-defense systems.

Swarm coordination concept:

flowchart TB
  SC[Swarm Controller]:::ctrl
  BRIEF[Initial mission brief]:::note
  ALPHA["Drone Alpha<br/>reconnaissance"]:::drone1
  BG["Drone Beta-Gamma<br/>perimeter hold"]:::drone1
  DO["Drone Delta-Omega<br/>strike / loiter"]:::drone2
  MESH[Mesh network<br/>between drones]:::note
  REDIST[Adaptive task<br/>redistribution]:::note
  SC --> BRIEF
  BRIEF --> ALPHA
  BRIEF --> BG
  BRIEF --> DO
  ALPHA --> MESH
  BG --> MESH
  DO --> MESH
  MESH --> REDIST
  classDef ctrl fill:#1e1b4b,stroke:#818cf8,color:#e0e7ff
  classDef note fill:#2e2410,stroke:#fbbf24,color:#fde68a
  classDef drone1 fill:#0f2a1c,stroke:#4ade80,color:#bbf7d0
  classDef drone2 fill:#2d1518,stroke:#f87171,color:#fecaca

This renders legacy air defense systems irrelevant — you cannot shoot down 1,000 units simultaneously.

US: Replicator Initiative

The US responded with the Replicator Initiative, announced by Deputy Secretary of Defense Kathleen Hicks in 2023 — targeting deployment of 1,000+ autonomous drones across multiple domains by 2025, led by the Defense Innovation Unit (DIU).

Focus: attritable drones — cheap, expendable assets, not high-value platforms like the F-35. The strategy was directly inspired by lessons from the Ukraine conflict.

Ukraine: The World’s First Live AI Drone Warfare Lab

The Ukraine conflict since 2022 became the first real-world test of AI-assisted drone warfare at scale:

The DJI case is a supply chain security wake-up call: consumer drones designed for aerial photography were modified into weapons. This has direct implications for every hardware vendor in the AI stack.

AI in Cyber Warfare

The battlefield isn’t only physical. AI has penetrated cyber warfare deeply:

GhostWriter (Russia) — large-scale disinformation campaign using AI-generated content to spread pro-Russia narratives across Poland, the Baltic states, and Ukraine before and during the conflict.

Volt Typhoon (China) — confirmed by FBI and CISA in 2024: Chinese APT group used AI-assisted reconnaissance to silently infiltrate critical US infrastructure, positioning for potential future disruption — operating undetected for years.

Most significant disclosure: In February 2024, Microsoft and OpenAI revealed that nation-state actors — including Russia (APT28/Forest Blizzard), China (Salmon Typhoon), Iran (Crimson Sandstorm), and North Korea — were using GPT and large language models to:

• Write attack code
• Research new vulnerabilities
• Generate more convincing phishing emails
• Translate technical documents for intelligence operations

“These actors were not using these tools to develop novel attack capabilities, but rather to improve their productivity and efficiency in existing operations.” — Microsoft Threat Intelligence, February 2024

For the Engineer: 4 Takeaways

:::note Key engineering implications from military AI deployments:

1. Dual-use is real — computer vision, LLMs, cloud security, autonomous systems: skills you build have defence applications. Understand the ethics before someone else decides for you.

2. AI supply chain = national security — DJI drones in Ukraine. Huawei equipment in critical infrastructure globally. Every component in an AI system has a geopolitical dimension.

3. AI safety scales with stakes — when AI is wrong in a chatbot, you get a weird answer. When AI is wrong in a targeting system, people die. Reliability, explainability, and human oversight are non-negotiable at this level.

4. Know what your work builds toward — Google engineers found out about Maven after it was running. Ask the question early.

   :::

Closing

Military AI is no longer a secret Pentagon pilot project. It is operational reality — tested in live conflict, funded in the billions, and being raced between great powers.

This series doesn’t take a position on whether this is good or bad. It is a factual field report for engineers who want to understand the world they’re building technology for.

Next in this series: Malaysia and Defence AI — do we have a strategy?