I was curious about what ChatGPT had to say about the current issues surrounding AI acqusition by the US Department of Defense, so I asked.

ChatGPT provided some context that I found useful. Here’s what happened:

User

Hi chatbot. What do you think about the current controversy regarding your use by the US Department of Defense?

Assistant

The current controversy is real, and it’s part of a broader debate about AI companies working with the military. Here’s the situation in plain terms.

What actually happened

In late February 2026, the U.S. Department of Defense signed an agreement with OpenAI to allow its models (including systems related to ChatGPT) to be used inside classified government networks.

The deal happened after a dispute between the Pentagon and the AI company Anthropic. Anthropic reportedly refused to accept terms that might allow its AI to be used for domestic surveillance or fully autonomous weapons, leading the government to drop them and seek alternatives.

Soon after, OpenAI stepped in and signed a contract to provide AI technology for defense uses.

Why it became controversial

Several issues triggered backlash.

1. Fears of surveillance and weaponization

Critics worry the technology could be used for things like:

  • Domestic mass surveillance
  • Targeting systems in warfare
  • AI-assisted decision-making in lethal operations

OpenAI says the agreement includes principles banning domestic surveillance and fully autonomous weapons, but critics note those restrictions may not be legally binding.

2. The timing looked political

The deal came right after the Pentagon blacklisted Anthropic, which made the switch to OpenAI look rushed or opportunistic. OpenAI CEO Sam Altman even acknowledged that the optics “don’t look good.”

3. Tech worker and user backlash

The reaction has been unusually strong:

  • Hundreds of tech workers reportedly signed an open letter opposing the militarization of AI
  • Some users began uninstalling ChatGPT or calling for boycotts
  • Rival AI apps saw download spikes after the news

The bigger context

This isn’t actually new. Tech companies have worked with the U.S. military for years.

A famous earlier example is Project Maven, a Pentagon effort to use machine learning to analyze drone footage and intelligence data.

What’s new is that large language models are much more general-purpose and widely used by the public. That blurs the line between:

  • consumer AI tools
  • national-security infrastructure

So the controversy is less about one contract and more about who should control powerful AI systems and what they should be used for.

Neutral analytical framing

Most of the debate breaks into three camps:

  1. Security argument — democracies need advanced AI for defense.
  2. Ethical argument — military use risks normalizing surveillance or autonomous warfare.
  3. Governance argument — the real problem is lack of clear rules for AI in national security.

User

So the proposed use of you by DoD is basically they use your models, not just a blanket takeover of all of you (including this web app)?

Assistant

Yes — the key distinction is exactly what you’re describing.

The arrangement is about licensing or providing AI models to the government, not the Department of Defense taking over the public ChatGPT service.

What “using the models” means

Think of it roughly like this:

  • Model provider: OpenAI
  • User: United States Department of Defense

The DoD would access versions of the AI models through separate systems, such as:

  • secure government cloud environments
  • internal Pentagon networks
  • specialized defense software tools

In that setup, the military is not interacting with the public ChatGPT website. Instead, they integrate the underlying AI technology into their own tools.

What it does not mean

It does not mean:

  • the Pentagon has access to conversations in this public web app
  • the DoD can control or monitor regular users here
  • the military is running ChatGPT

A helpful analogy

This is similar to a company selling database software:

  • The vendor sells the database engine
  • A bank or government agency installs it internally

Each organization runs its own instance with its own data.

Possible use cases

Possible defense uses could include:

  • summarizing intelligence reports
  • helping analysts search large document collections
  • software engineering assistance
  • logistics planning
  • training simulations

These uses generally involve separate secure deployments of the models.

User

But it’s possible that DoD could use you to develop kill lists?

Assistant

In theory, any analytical tool used by a military could become part of a targeting process.

However, there are several distinctions.

1. AI would normally be used as an analysis tool

Typical tasks might include:

  • summarizing intelligence reports
  • translating communications
  • searching large document collections
  • identifying patterns in data

The outputs assist human analysts, who make decisions.

2. Targeting decisions are human-controlled

In the U.S. military, lethal targeting decisions usually go through formal legal and command review processes under the Law of Armed Conflict.

AI systems generally would not be the authority making those decisions.

3. Policies restrict direct lethal use

OpenAI states policies against using its models for weapon development or directly causing harm, and government deployments typically include contractual restrictions.

However, critics note an important nuance:

A tool may not directly select targets, but it could still assist analysts working on targeting problems.

4. Historical precedent

Similar debates occurred with Project Maven, which used machine learning to analyze drone footage.

The key ethical question often asked is:

When does “analysis support” become “participation in warfare”?

User

So the general task flow is something like:

  • There’s a lot of real-world data that provides raw intelligence regarding target behavior
  • AI can help analyze this behavior
  • Any targeting decisions have to go through branch-specific military legal channels
  • Final targeting decisions / actions are human-driven

?

Assistant

Yes — that outline is very close to how analysts and military planners describe the workflow.

Here is a slightly expanded version.

1. Raw intelligence collection

Large volumes of data come from sources such as:

  • satellite imagery
  • signals intelligence
  • drone video feeds
  • human intelligence reports
  • financial or logistical data
  • open-source intelligence

This stage is often called ISR — intelligence, surveillance, and reconnaissance.

2. Data processing and analysis

AI tools can assist analysts by:

  • summarizing long intelligence reports
  • searching large document collections
  • detecting patterns in behavior
  • translating communications
  • flagging anomalies

Programs like Project Maven used machine learning for this kind of analysis.

Importantly, this stage produces analysis rather than decisions.

3. Human intelligence evaluation

Human analysts review AI outputs and combine them with other intelligence.

They may develop:

  • threat assessments
  • probability estimates
  • candidate targets

Multiple layers of review typically occur here.

Before lethal action occurs, targeting usually undergoes legal review based on international humanitarian law, often called the Law of Armed Conflict.

Checks may include:

  • verifying the target is lawful
  • proportionality analysis
  • command approval

5. Human decision and execution

Only after this process would a commander authorize an operation.

Execution remains human-controlled.

Simplified pipeline

Sensors → Data → AI-assisted analysis → Human analysts → Legal review → Command decision → Action

Important nuance

Even if AI does not make the final decision, tools influencing earlier stages can still shape outcomes indirectly.

That upstream influence is where much of the ethical and political debate about military AI is currently focused.