00:00:00
The US military is no longer testing
00:00:01
generative AI. It's deploying it. In
00:00:05
2025, Marines in the Pacific use chat
00:00:07
GPT style systems during real operations
00:00:10
to analyze surveillance, flag threats,
00:00:13
and assist in decision-making. This
00:00:15
marks the shift into phase 2 of military
00:00:17
AI, where language models aren't just
00:00:20
processing data, but actively shaping
00:00:22
strategy. In this video, we'll break
00:00:24
down what's really happening, what
00:00:26
systems are in play, and how this
00:00:28
changes the rules of modern warfare. And
00:00:31
by the end, you'll also see what comes
00:00:33
next, including risks no one's fully
00:00:35
prepared for yet. What's actually
00:00:38
happening on the ground? So, how exactly
00:00:40
are these generative AI systems being
00:00:42
used in field deployments? In one
00:00:44
instance, Marines were able to ask the
00:00:47
system specific questions like, "What
00:00:49
are the most recent drone sightings in
00:00:51
this sector?" or summarize the satellite
00:00:54
report for enemy movement in the last 12
00:00:57
hours. The AI would respond in seconds
00:00:59
with synthesized outputs compiled from
00:01:02
raw data feeds, radar logs, and prior
00:01:04
reports, all in plain language. Instead
00:01:07
of combing through multiple dashboards,
00:01:09
or intelligence memos, they had answers
00:01:11
in real time. The system didn't just
00:01:13
provide summaries. It also flagged
00:01:15
anomalies, identified potential threats,
00:01:18
and in some cases, suggested follow-up
00:01:20
actions. Though importantly, human
00:01:23
officers still made the final call.
00:01:25
According to a recent report by Rand
00:01:27
Corporation, military commanders are
00:01:29
becoming more reliant on AI generated
00:01:31
insights for operational level
00:01:33
decisions, not just tactical support.
00:01:36
This means AI tools are being used to
00:01:38
recommend troop movements, identify
00:01:41
vulnerabilities in terrain, or even
00:01:43
prioritize surveillance zones. The
00:01:45
models in use are based on large
00:01:47
language model architectures similar to
00:01:49
the technology behind OpenAI's GPT4, but
00:01:53
they are fine-tuned on military data
00:01:55
sets. And Palanteer and Microsoft have
00:01:58
all been developing custom models
00:02:00
specifically for defense use cases. In
00:02:03
March 2025, OpenAI confirmed a defense
00:02:06
focused partnership with Andrew to
00:02:08
integrate generative models into
00:02:10
battlefield systems, marking a major
00:02:12
shift for the company, which previously
00:02:14
avoided military contracts. These
00:02:16
systems are being tested in secure cloud
00:02:18
environments, often hosted on Azure
00:02:21
government or classified networks, and
00:02:23
are governed under Department of Defense
00:02:25
AI policies drafted in 2023 and updated
00:02:29
again under Trump's administration in
00:02:31
February 2025. As of April, these tools
00:02:34
remain in the assisted category. They're
00:02:36
not authorized to make autonomous
00:02:38
decisions or initiate actions, but
00:02:40
they're deeply embedded in the analysis
00:02:42
and advisory layers of modern combat
00:02:44
operations. The illusion of human in the
00:02:46
loop. The Department of Defense often
00:02:49
reassures the public with one key
00:02:51
phrase. There will always be a human in
00:02:53
the loop. It's meant to ensure that no
00:02:56
AI system will have full control over
00:02:58
life or death decisions without human
00:03:00
intervention. But experts are
00:03:02
questioning whether this concept still
00:03:04
holds weight, especially as AI models
00:03:06
become increasingly complex and
00:03:08
fastmoving. According to Heidi Cloff, a
00:03:11
safety engineer and the current chief AI
00:03:13
scientist at the AI Now Institute, the
00:03:16
idea of human in the loop can be
00:03:18
misleading. In her recent statement to
00:03:20
MIT Technology Review, she explained
00:03:22
that when AI models synthesize data from
00:03:25
thousands of different sources, it
00:03:27
becomes almost impossible for a human to
00:03:29
properly audit the outcome in real time.
00:03:32
The AI's reasoning is based on so many
00:03:34
variables that a human would need hours,
00:03:37
if not days, to validate what the AI did
00:03:39
in seconds. This creates a fundamental
00:03:42
tension. On paper, a human is always
00:03:45
approving the final decision. But in
00:03:47
practice, that human may be relying
00:03:49
almost entirely on the AI's
00:03:51
recommendation because there's simply no
00:03:53
time or ability to verify every detail
00:03:56
in fast-paced environments like
00:03:58
battlefield command centers. This
00:04:00
problem only scales as AI becomes more
00:04:02
embedded across layers of military
00:04:04
infrastructure. In 2024, the Defense
00:04:07
Innovation Board warned that reliance on
00:04:09
blackbox models without transparency
00:04:12
would lead to scenarios where commanders
00:04:14
may believe they're in control, but
00:04:16
they're really just rubber stamping what
00:04:18
the model outputs. Artificial
00:04:19
intelligence is breaking how we handle
00:04:21
classified information. Generative AI is
00:04:24
disrupting how the military determines
00:04:26
what should be classified.
00:04:28
Traditionally, intelligence was manually
00:04:30
tagged by analysts. Now, AI systems can
00:04:33
process vast amounts of unclassified
00:04:35
data like satellite images and news
00:04:37
reports and synthesize insights that
00:04:40
would typically be considered
00:04:41
classified. This is known as
00:04:43
classification by compilation. Chris
00:04:46
Motton of Rand notes that there's still
00:04:48
no clear framework for handling these AI
00:04:51
generated outputs. Tools like those from
00:04:53
Palunteer and Microsoft aim to automate
00:04:56
classification using probabilistic
00:04:58
models, some even trained on sensitive
00:05:00
data sets. But without a standardized
00:05:02
protocol, oversight remains
00:05:04
inconsistent. The volume of data only
00:05:07
makes the challenge harder. Drones,
00:05:09
satellites, and battlefield sensors are
00:05:11
generating terabytes daily, and AI is
00:05:14
producing real-time analyses from that
00:05:16
stream. Each summary could potentially
00:05:18
contain sensitive insights. Yet no
00:05:21
current system reliably flags what needs
00:05:23
protection. With the pace of AI
00:05:26
generation far outstripping human
00:05:28
review, the Pentagon is now grappling
00:05:30
with a core question. When intelligence
00:05:33
is machine created, who decides what
00:05:35
stays secret and what doesn't? How far
00:05:37
up the chain will AI climb? In 2017,
00:05:41
military AI was largely confined to
00:05:44
tactical applications. The most
00:05:46
well-known example was Project Maven,
00:05:48
which used computer vision to detect
00:05:50
objects like people, vehicles, and
00:05:52
buildings in drone footage. These
00:05:54
systems provided support, but stopped
00:05:56
short of interpreting situations or
00:05:59
suggesting decisions. That's no longer
00:06:01
the case. As of 2025, the role of AI has
00:06:05
expanded dramatically, now pushing into
00:06:07
operational decision-making, where it
00:06:09
helps shape the outcomes of real-time
00:06:11
military missions. According to a March
00:06:14
2025 report by the Center for Security
00:06:17
in emerging technology at Georgetown
00:06:19
University, there's been a significant
00:06:21
increase in AI being used to assist
00:06:23
commanders during live operations. This
00:06:26
includes suggesting optimal troop
00:06:27
movements, identifying emerging threats,
00:06:30
and analyzing battlefield conditions to
00:06:32
inform next steps. The report noted that
00:06:35
AI tools are no longer limited to
00:06:37
analysis. They're directly influencing
00:06:39
military strategy, particularly in
00:06:41
complex environments with limited
00:06:43
response time. Looking ahead, the next
00:06:45
phase includes the adoption of agentic
00:06:47
AI systems that can not only respond to
00:06:50
commands but initiate task independently
00:06:53
and personalized AI models that adapt to
00:06:55
the preferences or patterns of
00:06:57
individual users. These technologies are
00:07:00
already in use in civilian sectors and
00:07:02
are being piloted for defense
00:07:03
applications. In October 2024, the Biden
00:07:07
administration issued a national
00:07:09
security memorandum on AI outlining
00:07:11
ethical standards and requiring human
00:07:13
oversight for military AI systems.
00:07:16
However, as of early 2025, the Trump
00:07:18
administration has called for fewer
00:07:20
restrictions, stating that innovation
00:07:22
and speed are key to maintaining an edge
00:07:25
over adversaries. This shift raises new
00:07:27
concerns about how high AI will climb in
00:07:30
the military chain of command and what
00:07:32
happens if those safeguards are no
00:07:34
longer enforced. Palunteer, Microsoft,
00:07:37
and the new AI arms race. As the
00:07:40
Pentagon deepens its use of generative
00:07:42
AI, private defense contractors are
00:07:44
competing for dominance in this rapidly
00:07:46
evolving space. Palunteer Technologies
00:07:49
has been at the forefront offering AI
00:07:52
platforms capable of automating
00:07:54
classification, flagging threats, and
00:07:57
analyzing intelligence at scale. Its
00:08:00
systems are designed to integrate
00:08:01
seamlessly with existing US military
00:08:04
data infrastructure and have already
00:08:06
been deployed in test environments to
00:08:08
support real-time battlefield decisions.
00:08:10
Microsoft is also playing a key role
00:08:13
through its Azure government cloud and
00:08:15
partnerships with defense agencies.
00:08:17
Microsoft has been developing generative
00:08:19
AI models trained on sensitive and
00:08:21
classified data. These models are
00:08:23
designed to support everything from
00:08:24
logistics to targeting and the company
00:08:27
has emphasized the importance of secure
00:08:29
training environments to prevent
00:08:31
unauthorized data leaks. In March 2025,
00:08:35
OpenAI entered the defense sector
00:08:37
through a partnership with Andreal
00:08:39
Industries, marking a notable shift from
00:08:41
its earlier public stance on military
00:08:43
use. Under this agreement, OpenAI's
00:08:46
models will be integrated into Andrew's
00:08:48
autonomous systems, bringing generative
00:08:50
AI closer to real-time battlefield
00:08:52
deployment. This growing reliance on
00:08:55
private firms introduces critical
00:08:56
challenges around accountability, data
00:08:59
governance, and the alignment of
00:09:01
corporate incentives with military
00:09:02
objectives. While these companies are
00:09:05
helping to accelerate innovation, their
00:09:08
expanding role raises important
00:09:09
questions about transparency and ethical
00:09:12
oversight in an increasingly automated
00:09:14
war environment. The bigger risk no
00:09:16
one's talking about. As AI systems gain
00:09:19
influence in military operations, one
00:09:21
central issue remains largely
00:09:23
unresolved. Who is accountable when
00:09:26
something goes wrong? Generative AI
00:09:29
excels at pattern recognition and rapid
00:09:31
synthesis, but it lacks human judgment,
00:09:34
contextual understanding, and moral
00:09:36
reasoning. Human rights organizations,
00:09:38
including Human Rights Watch and the
00:09:40
Campaign to Stop Killer Robots, have
00:09:43
repeatedly warned that relying on AI to
00:09:45
assist with or recommend lethal actions
00:09:48
risk detaching accountability from human
00:09:50
actors. A misinterpretation of satellite
00:09:53
imagery, a flawed data set, or a biased
00:09:56
input could lead an AI system to
00:09:58
recommend a strike on the wrong target.
00:10:00
Even if a human signs off, did they
00:10:02
fully understand the AI's rationale? And
00:10:05
if not, who is to blame if civilians are
00:10:07
harmed? This isn't just a technical
00:10:09
flaw. It's a structural gap in modern
00:10:12
warfare. Without clear lines of
00:10:14
responsibility, AI generated decisions
00:10:16
could lead to catastrophic mistakes with
00:10:19
no clear accountability. What happens
00:10:21
next? The military's push into
00:10:23
generative AI is not slowing down. With
00:10:26
rising geopolitical tensions and rapid
00:10:28
advances from foreign adversaries,
00:10:30
particularly China and Russia, the US is
00:10:33
under pressure to maintain its
00:10:35
technological edge. This urgency is
00:10:37
fueling heavy investment from both the
00:10:39
government and the private sector. With
00:10:41
billions of dollars being poured into
00:10:43
dualuse AI research, battlefield
00:10:46
applications, and strategic automation.
00:10:49
But this rapid development raises urgent
00:10:51
questions that remain unanswered. How
00:10:53
can real human oversight be guaranteed
00:10:56
in high-speed decision-making
00:10:57
environments? Should AI ever be allowed
00:11:00
to assist in decisions that could lead
00:11:02
to loss of life? And how do we avoid the
00:11:04
slow, unchecked creep toward fully
00:11:07
autonomous weapon systems, a path that
00:11:09
has long been opposed by international
00:11:11
watchdogs? Phase 2 of military AI isn't
00:11:14
just a technical milestone. It's a
00:11:17
structural transformation. one that
00:11:19
redefineses how power is exercised, how
00:11:22
wars are fought, and who gets to decide
00:11:24
the cost of those decisions. If you've
00:11:26
made it this far, let us know what you
00:11:28
think in the comment section below. For
00:11:31
more interesting topics, make sure you
00:11:33
watch the recommended video that you see
00:11:34
on the screen right now. Thanks for
00:11:36
watching.
