The Negative 'Learning Curve' of Autonomous Weapons: The Reality of Anduril's Failures and the Risks Investors Overlook
Anduril is an up-and-coming defense company that has achieved a multi-billion dollar valuation in just a few years, positioning drones and autonomous weapons as the protagonists of next-generation warfare. However, according to recent reports, it is facing serious problems such as 'repeated failures in experiments' and 'unreliability on the battlefield,' exposing not only the potential of autonomous weapons but also their limitations and dangers. In this article, we examine 'speed-oriented weapons development' and the 'structural risks of autonomous weapons' through specific failure cases.
1. What is Anduril / What were they aiming for?
Anduril was founded in 2017 and is a company that handles autonomous or remotely operated weapon groups, such as unmanned aerial vehicles (UAV/UAS), counter-drone systems, and command and control software.
Representative products include the tube-launched fixed-wing drone Altius, the small helicopter-type reconnaissance Ghost drone, the Anvil counter-drone system, and the Fury (YFQ-44) combat drone that will coordinate with future manned aircraft.
Anduril has marketed itself on 'integrated control via software,' 'force projection through the deployment of large numbers of inexpensive drones,' and 'rapid development cycles.' For example, the concept is that through the company's command and control software platform, Lattice, the functionality of a vast number of drones can be improved at the same speed as code updates.
However, this 'speed and scale-oriented' approach places a heavy burden on the systems, and failures in actual combat and testing have occurred one after another.
2. Recent major failure cases — Stumbling in both testing and actual combat
2-1. Runaway and shutdown of unmanned boats (US Navy exercise)
In May 2025, during a Navy exercise off the coast of California, when they attempted to deploy over 30 unmanned drone boats, more than 12 'ignored instructions' and stopped. Although they stopped navigating due to a so-called fail-safe, it was considered a 'significant operational and safety risk,' and Navy officials strongly criticized Anduril's software operations in a subsequent report. It is said that there was even an unusual mention pointing out 'continuous operational safety violations.' Such control failures are anomalies that would almost never happen in conventional weapons development, where it is not considered acceptable just because 'the worst-case scenario was avoided as a result' rather than going as planned.
2-2. Ground test engine damage of the unmanned combat aircraft Fury
Anduril is developing the Fury in response to the US Air Force's 'manned-unmanned teaming (MUM-T)' concept, but during ground tests in the summer of 2025, a 'nail was sucked into the air intake, causing engine damage.' As a result, the first flight was postponed. In the fighter class, where reliability equivalent to that of a manned aircraft is required, such a basic trouble as engine damage speaks to the high hurdle of 'actual deployment equals stable operation.'
2-3. 22-acre wildfire in Oregon, USA, caused by malfunction of the Anvil counter-drone device
In an August 2025 test, the Anvil system accidentally caused a fire, leading to the burning of 22 acres (over 9 hectares) of terrain. It was a large-scale fire that required the mobilization of fire departments, and it is said to have been an unexpected accident for the drone developers. Such 'uncontrollable behavior' that leads to serious accidents is always lurking in autonomous weapons.
2-4. 'Lack of reliability' of drones deployed in actual combat — The setback of Altius in Ukraine
Anduril is said to have provided about 100 Altius drones to Ukraine, but according to frontline troops on the ground, the loitering-type Altius drones failed to reach their targets, crashed, and missed their hits one after another, with reports that their use stopped in 2024. While Anduril claims that 'failures are part of weapons development' and that they are 'already working on improvements through software revisions,' 'weapons that cannot hit' in actual combat are not just 'experiments' but involve life-or-death issues.
3. Why are failures occurring one after another — The curse of speed and reliance on autonomy
3-1. Rapid development cycles and a culture that 'assumes failure'
In traditional defense companies, it was normal for testing, verification, manufacturing, and delivery to take years to decades. In contrast, Anduril uses a 'Silicon Valley style' of repeating 'development, testing, revision, and deployment' in units of months. Therefore, it is a development method that assumes failure, but in the field of 'weapons,' which involves lives, 'it's good that no one died in the test' is by no means sufficient. In a WSJ report, the company even explained, 'We do fail… a lot.'
3-2. The 'black box nature' and unpredictability of autonomy/AI
In the first place, autonomous weapons/AI weapons carry the risk of not moving as designed. Recent academic research points out that no matter how highly designed Lethal Autonomous Weapons Systems (LAWS) are, they harbor fundamental vulnerabilities such as 'decision-making black boxes' and 'unpredictable emergent behavior.' These accidents and failures can be seen as the realization of that very danger.
4. Ethical and safety risks that must be faced alongside the potential of autonomous weapons
The potential benefits brought by autonomous weapons—mass deployment, cost efficiency, and rapid response—are certainly attractive. However, the Anduril case demonstrates how fatal the 'uncertainty of reliability,' 'black-box nature,' and 'risk of loss of control' of autonomous weapons can be when the conditions of 'startup speed-supremacy' and 'short-term mass production and deployment' are added.
Furthermore, accidents caused by malfunctions are not merely technical failures; they have the potential to cause loss of life, civilian harm, and issues under international law.
Conclusion
Anduril's series of failures is a realistic and stern warning against the illusion that 'autonomous weapons equal the future of military power.'
Tactical drones, unmanned combat aerial vehicles, anti-drone devices... the concept of mass-deploying these as 'software-controlled weapon systems' may be theoretically advanced and innovative. However, for actual operation, isn't the cautious process adopted by traditional weapons development—'caution over speed,' 'verification over code updates,' and 'safety management over innovation'—indispensable?
If we are to discuss the future of autonomous weapons development, we must once again calmly consider the 'possibilities' and 'dangers' based on concrete examples.