.toc-list { position: relative; } .toc-list { overflow: hidden; list-style: none; } .gh-toc .is-active-link::before { background-color: var(-ghost-accent-color); /* Determines the highlight color of the table of contents based on the highlight color set in Ghost Admin */ } .gl-toc__header { align-items: center; color: var(–foreground); cursor: cursor; Display: Flex; gap: 2rem; set-content:space-between; padding: 1 rim; width: 100%; } .gh-toc-title { font-size: 15 pixels! important; font-weight: 600 !important; character spacing: .0075rem; line-height: 1.2; margin: 0; text-transform: uppercase; } .gl-toc__icon { transition: easily convert .2s to outside; } .gh-toc me { color: #404040; font-size: 14px; line-height: 1.3; margin-bottom: .75rem; } .gh-toc { display: none; } .gh-toc.active { display: block; } .gl-toc__icon svg{ transition: easily convert 0.2s abroad; } .gh-toc.active + .gl-toc__header .gl-toc__icon .rotated{ transform: rotate(180deg); } .gl-toc__icon .rotated{ transform: rotate(180°); } .gh-toc-container-sidebar{ display: none; } .gh-toc-container-content{ display: block; width: 100%; } a.toc-link{ background-image: none! important; } .gh-toc-container-content .toc-list-item{ margin-left: 0 ! important; } .gh-toc-container-content .toc-list-item::marker{ content: none; } .gh-toc-container-content .toc-list{ padding: 0 ! important; Margin: 0! important; } media screen only and (min-width: 1200px) { .gh-sidebar-wrapper{ margin: 0; Position: sticky. Top: 6 Rim. left: calculate(((100vw – 928px)/ 2 ) – 16.25rem – 60px); z-index: 3; } .gh-sidebar { align-self: flex-start; background-color: transparent; flex direction: column; Network area: tok; max height: calculate(100vh – 6rem); Width: 16.25 rem; z-index: 3; Position: sticky. top: 80px; } .gh-sidebar:before { -webkit-backdrop-filter: Blur(30px); background-filter: blur(30px); background-color:hsla(0, 0%, 100%, .5);; border-radius: .5rem; content: “”; display: block; height: 100%; left: 0; Position: absolute; top: 0; width: 100%; z-index: -1; } .gl-toc__header { cursor: default; Elastic shrinkage: 0; cursor-events:none; } .gl-toc__icon { display: none; } .gh-toc { display: block; Flex: 1; overflow p: auto; } .gh-toc-container-sidebar{ display: block; } .gh-toc-container-content{ display: none; } } ))>
Titles
On February 13, the Wall Street Journal published something that had never been made public before:… The Pentagon used Claude AI from Anthropic During the January raid that captured Venezuelan leader Nicolas Maduro.
Claude’s deployment came about through Anthropic’s partnership with Palantir Technologies, whose platforms are widely used by the Department of Defense, she said.
Reuters tried to independently verify the report, but was unable to do so. Anthropic declined to comment on specific operations. The Ministry of Defense declined to comment. Palantir said nothing.
But the Wall Street Journal report revealed other details.
Sometime after the January raid, an Anthropic employee reached out to someone at Palantir and asked a direct question: How was Claude actually being used in that operation?
The company that built the model and signed the $200 million contract had to ask someone else what its software did during a military attack on the capital.
These details tell you everything about where we actually stand in terms of AI governance. It also tells you why the “human in the loop” stopped being a guarantee of safety somewhere between the signing of the contract and Caracas.
How big was the operation?
Calling this a secret extraction misses what actually happened.
Delta Force raided multiple targets throughout Caracas. More than 150 aircraft participated. Air defense systems were suppressed before the first boots hit the ground. Airstrikes hit military targets and air defences, and electronic warfare assets were transferred to the area. According to Reuters.
Cuba later confirmed the deaths of 32 of its soldiers and intelligence personnel and declared two days of national mourning. The Venezuelan government stated that the death toll reached about 100 people.
Two sources told Axios That Claude was used during the active operation itself, although Axios noted that it could not confirm the exact role Claude played.
What Claude could have actually done
To understand what could happen, you need to know one technical thing about how CLOUD works.
The Anthropic API is stateless. Each call is independent, meaning you send a text message, receive a text message back, and that interaction ends. There is no persistent memory or cloud running constantly in the background.
It’s less like a mind and more like a very quick advisor that you call every thirty seconds: you describe the situation, they give you their best analysis, you hang up, and you call back with new information.
This is the API. But that says nothing about the systems Palantir has built on top of it.
You can design a proxy ring that feeds real-time information to the cloud continuously. You can create workflows where Claude’s output triggers the next action with minimal latency between recommendation and execution.
Testing these scenarios myself
To understand what this actually looks like in practice, I tested some of these scenarios.
Every 30 seconds. Indefinitely.
Stateless API. It doesn’t have to be a sophisticated military system built on an API.
What this might look like when deployed:
Intercepted communications in Spanish were sent to Claude for interpretation and pattern analysis across hundreds of messages simultaneously. Satellite images are processed to determine vehicle movements, troop locations, or infrastructure changes, with updates every few minutes as new images arrive.
Or compile real-time intelligence from multiple sources — signal intercepts, human intelligence reports, electronic warfare data — and compress it into actionable briefings that would take analysts hours to produce manually.
I practiced scenarios. Distributed in Caracas.
None of this requires Claude to “decide” anything. It’s all analysis and synthesis.
But when you compress a four-hour intelligence cycle into minutes, and that analysis feeds directly into operational decisions made in the same compressed timescale, the distinction between “analysis” and “decision-making” begins to break down.
Because this is a secret network, no one outside this system knows what has actually been built.
So when someone says “Cloud can’t run a standalone process” – they are probably right regarding the API level. Whether they are right about the level of publication is a different question entirely. One that no one can currently answer.
The gap between autonomy and meaningfulness
The hard limit of anthropology is autonomous weapons, systems that decide to kill without human consent. This is a real line.
But there is a huge amount of ground between “autonomous weapons” and “meaningful human oversight.” Think about what this means practically for a commander in an active operation. CLOUD aggregates intelligence across volumes of data that no analyst can hold in his head. It compresses what was previously a four-hour briefing course into minutes.
This took 3 seconds.
It shows patterns and recommendations faster than any human team can produce them.
Technically, a human agrees to everything before taking any action. human in this process. But the process is now moving so quickly that it is impossible to assess what is in it in fast-paced scenarios such as a military attack. When Claude creates an intelligence summary, this summary becomes an input for the next decision. Because Claude can produce these summaries much faster than humans can process them, the pace of the entire process accelerates.
You can’t slow down to think carefully about a recommendation when the situation you’re describing is already three minutes old. Information has been transmitted. The next update is already arriving. The ring keeps getting faster and faster.
90 seconds to make a decision. This is what the ring looks like from the inside.
The requirement for human consent is there but the ability to meaningfully evaluate what you are consenting to is not.
And structurally it gets worse as AI gets better, because better AI means faster synthesis, shorter decision windows, and less time to think before acting.
Pentagon and Claude arguments
The Pentagon wants access to artificial intelligence models For any use case that complies with US law. Their position is basically: the usage policy is our problem, not yours.
But Anthropic wants to keep certain prohibitions in place – the lack of fully autonomous weapons and the prohibition of mass domestic surveillance of Americans.
After the Wall Street Journal broke the story, a senior administration official told Axios that their partnership/agreement is under review and here’s why the Pentagon said:
“Any company that would jeopardize the operational success of our warfighters in the field is a company we need to re-evaluate.”
But ironically, Anthropic is currently the only approved commercial AI model for some classified DoD networks. Although OpenAI, Google, and xAI are all actively discussing access to those systems with fewer restrictions.
The real battle behind the arguments
In hindsight, Anthropics and the Pentagon may miss the whole point and think that political languages might solve the problem.
Contracts can require human approval at every step. But that doesn’t mean a human has enough time, context, or cognitive bandwidth to actually evaluate what they’re agreeing to. That gap between the human who is technically in the loop and the human who is actually able to think clearly about what is in it is where the real danger lies.
Rogue AI and autonomous weapons will likely be the next set of arguments.
The discussion today should be: Can we call it “supervised” when you put a system that processes large amounts of information faster than humans up a human chain of command?
Final thoughts
In Caracas, in January, with 150 planes and real-time briefs and decisions being made at operational speed, we don’t know the answer to that.
Neither does Anthropy.
But soon, with fewer restrictions and more models on these underground networks, we’ll all find out.
All claims in this piece are sourced from public reports and documented specifications. We have no non-public information about this process. Sources: Wall Street Journal (February 13), Axios (February 13, February 15), Reuters (January 3, February 13). Casualty figures from the official Cuban government statement and the Venezuelan Ministry of Defense. API architecture from platform.claude.com/docs. Contract details from Anthropic’s August 2025 press release. “Vision in Use” quote from Axios (February 13).
