Anthropic Palantir
Anthropic pledged their AI would never support autonomous weapons or surveillance. Now their AI is in classified military operations. No community input was sought on whether AI should be used in military decision-making.
In 2025, Anthropic became the first AI company cleared for classified military operations. Through a partnership with Palantir—the data analytics company that's been in the defense intelligence game since its CIA-funded inception—Anthropic's Claude AI is now operating in environments where the stakes aren't customer service tickets or code reviews.
The stakes extend far beyond commercial applications.
The Promise
Let's start with what Anthropic said they would never do:
Anthropic's Stated Commitments:
"We will not sell our AI for military applications"
"We will not support autonomous weapons systems"
"We will not assist with surveillance that violates civil liberties"
"We will maintain responsible scaling policies"
The Marketing:
Anthropic positioned itself as the "safe" AI company
Founded by former OpenAI employees concerned about AI safety
Claude was designed with "constitutional AI" principles
The company emphasized ethical AI development
The Reality:
Palantir partnership announced for defense and intelligence
First AI company cleared for classified operations
Claude now operating in military decision-support systems
The line between "support" and "weapons" is conveniently blurry
Anthropic's pledge against military applications now reads differently: they will not sell AI directly for military use, but they will partner with a company whose primary business is defense intelligence.
The Palantir Connection
Understanding this story requires understanding Palantir:
What Palantir Does:
Builds data analytics platforms for intelligence agencies
Works with CIA, NSA, FBI, and military branches
Has been involved in military targeting systems
Was instrumental in finding Osama bin Laden
Provides "predictive analytics" for defense operations
Palantir's History:
Founded with CIA venture capital (In-Q-Tel)
Named after the "seeing stones" in Lord of the Rings
Has been controversial for surveillance applications
CEO Alex Karp openly embraces defense contracts
The Partnership:
Anthropic provides the AI language model
Palantir provides the defense infrastructure
Together, they offer "AI-powered intelligence analysis"
The system operates in classified environments
This arrangement is functionally equivalent to a pacifist handing bullets to a gun manufacturer and claiming no involvement in shooting.
What "Classified Operations" Means
When Anthropic says their AI is in "classified operations," here's what that could mean:
Intelligence Analysis:
Processing intercepted communications
Analyzing satellite imagery
Identifying targets from surveillance data
Predicting enemy movements
Decision Support:
Recommending military actions
Assessing threat levels
Prioritizing targets
Allocating resources
The Autonomy Question:
Is Claude making recommendations that humans follow blindly?
At what point does "decision support" become "decision making"?
How much human oversight exists in time-critical operations?
Who's accountable when AI-assisted decisions go wrong?
The Classified Problem:
We don't know exactly what the AI is doing
The operations are secret by definition
Oversight is limited to cleared personnel
Public accountability is impossible
The EFF's Warning
The Electronic Frontier Foundation has raised specific concerns:
EFF's Position:
"Tech companies shouldn't be bullied into surveillance"
Military contracts create structural incentives
Classified operations prevent public oversight
The national security apparatus exploits AI companies
The Pressure Dynamic:
Government agencies offer massive contracts
Refusing means competitors accept the work
"If we don't do it, someone less careful will"
The race to the bottom accelerates
The Structural Problem:
AI companies need revenue to survive
Defense contracts are lucrative and stable
Ethical commitments are expensive to maintain
Market pressures favor compromise
The "Dual Use" Defense
Anthropic and similar companies often use the "dual use" defense:
The Argument:
"Our AI has many applications, not just military"
"We can't control how customers use our technology"
"The same AI that helps doctors can help soldiers"
"We provide tools, not applications"
The Rebuttal:
When you partner with a defense contractor, you know the use case
"Dual use" doesn't mean "we have no responsibility"
Classified operations aren't ambiguous about their purpose
The partnership structure implies knowledge and consent
The Reality:
Anthropic chose to partner with Palantir specifically
They sought and obtained classified clearance
They knew exactly what they were getting into
The "dual use" defense is, willful blindness
The Consent Problem
Here's where this connects to our core mission: there was no public input:
Who Wasn't Consulted:
The public, who will live with the consequences of AI warfare
The communities targeted by AI-assisted operations
The soldiers who must trust AI recommendations
The future generations who will inherit this technology
The Democratic Deficit:
No public debate about AI in military operations
No congressional oversight of specific AI deployments
No international agreements on AI warfare
No consent from affected populations
The Precedent:
Once AI is in the military, it's hard to remove
Other countries will develop their own military AI
The arms race accelerates
The technology proliferates
The Broader AI Military Landscape
Anthropic isn't alone, but they are significant:
Other AI Military Contracts:
OpenAI: Removed "military" from prohibited uses in January 2024
Google: Project Maven controversy, then quietly resumed defense work
Microsoft: Extensive military contracts, including HoloLens for soldiers
Amazon: CIA cloud contracts, facial recognition for law enforcement
The Pattern:
Company pledges not to work with military
Company faces financial pressure
Company finds "responsible" way to accept military contracts
Company claims their involvement makes things safer
Repeat
The Exception:
Some companies have maintained their commitments
But they're increasingly rare and financially disadvantaged
The market rewards compromise
Ethics become a luxury good
What This Means for AI Safety
Anthropic's military partnership has implications beyond this one contract:
The Safety Paradox:
Anthropic was founded to develop safe AI
Military applications are among the most dangerous uses
The company now enables exactly what it feared
"Constitutional AI" principles meet real-world compromise
The Trust Problem:
If Anthropic compromises on military use, what else will they compromise?
Can any AI company's safety commitments be trusted?
Will "responsible scaling" policies actually be enforced?
Who watches the watchmen?
The Arms Race:
AI in military creates pressure for more AI in military
Defensive AI requires offensive AI to test against
The escalation dynamic is self-reinforcing
International agreements lag behind technology
Defending Yourself
As a Citizen:
Demand congressional oversight of AI military applications
Support international agreements on AI warfare
Advocate for transparency in defense AI contracts
Push for public debate before deployment
As a Consumer:
Consider which AI companies align with your values
Support organizations like EFF fighting for oversight
Be skeptical of "responsible AI" marketing claims
Remember that corporate pledges can be broken
As a Citizen:
Recognize that AI in warfare affects everyone
Understand that "classified" doesn't mean "not happening"
Question the inevitability narrative around military AI
Insist that consent matters, even for technology
Summing Up
Anthropic promised their AI would never be used for military applications. Now it's in classified operations through Palantir. The company that was supposed to be the "safe" AI option has joined the military-industrial complex.
They didn't ask if AI should be in the business of war. The contract was signed regardless.
The question is not whether every AI company has a price. The question is whether self-imposed safety commitments mean anything when financial incentives push the other way.
—-
This article is part of our ongoing coverage of AI consent violations. For more on military AI and surveillance, see the Pentagon's trillion-dollar consent problem.