"The Maven Smart System is the platform that came out of those exercises, and it, not Claude, is what is being used to produce “target packages” in Iran.
-
"The Maven Smart System is the platform that came out of those exercises, and it, not Claude, is what is being used to produce “target packages” in Iran. There are real limits to what a civilian like myself can know about this system, and what follows is based on publicly-available information, assembled from Palantir product demos, conferences, as well as instructional material produced for military users. But we can know quite a bit. The interface looks like a tacticool, dark mode send-up of enterprise software paired with the features of geospatial application like ArcGIS. What the operator sees are either maps with GIS-like overlays or a screen organized like a project management board. There are columns representing stages of the targeting process, with individual targets moving across them from left to right, as in a Kanban board.
Before Maven, operators worked across eight or nine separate systems simultaneously, pulling data from one, cross-referencing in another, manually moving detections between platforms to build a targeting case. Maven consolidated and orchestrated all of these behind a single interface. Cameron Stanley, the Pentagon’s chief digital and AI officer, called it an “abstraction layer,” a common term in software engineering, meaning a system which hides the complexity underneath it.16 Humans run the targeting and the ML systems underneath produce confidence intervals. Three clicks convert a data point on the map into a formal detection and move it into a targeting pipeline. These targets then move through columns representing different decision-making processes and rules of engagement. The system evaluates factors and presents ranked options for which platform and munition to assign, what the military calls a Course of Action. The officer selects from the ranked options, and the system, depending on who is using it, either sends the target package to an officer for approval or moves it to execution.
The AI underneath the interface is not a language model, or at least the AI that counts is not. The systems that detect targets in satellite imagery, fuse data from radar and drone footage, and track objects across multiple intelligence sources are computer vision and sensor fusion.17 They predate large language models by years. Neither Claude nor any other LLMs detects targets, processes radar, fuses sensor data, or pairs weapons to targets. LLMs are late additions to Palantir’s ecosystem;they were added in late 2024, years after the core system was operational, “AIP” was added as a natural language layer that summarizes documents or constructs and answers queries.18 When Anthropic was blacklisted, the Pentagon signed a replacement contract with OpenAI within hours. Replacing one language model with another is often just a simple configuration change, all you really have to do is change the API endpoint.
The language model was never what mattered about this system. What mattered was what Maven did to the process: it consolidated the systems, compressed the time, and reduced the people. That is not a new idea. The United States military has been trying to close the gap between seeing something and destroying it for as long as that gap has existed, and every attempt has produced the same failure. Maven may not even be the most extreme case."
Kill Chain
On the automated bureaucratic machinery that killed 175 children
(artificialbureaucracy.substack.com)
-
P pixelate@tweesecake.social shared this topic
-
"The Maven Smart System is the platform that came out of those exercises, and it, not Claude, is what is being used to produce “target packages” in Iran. There are real limits to what a civilian like myself can know about this system, and what follows is based on publicly-available information, assembled from Palantir product demos, conferences, as well as instructional material produced for military users. But we can know quite a bit. The interface looks like a tacticool, dark mode send-up of enterprise software paired with the features of geospatial application like ArcGIS. What the operator sees are either maps with GIS-like overlays or a screen organized like a project management board. There are columns representing stages of the targeting process, with individual targets moving across them from left to right, as in a Kanban board.
Before Maven, operators worked across eight or nine separate systems simultaneously, pulling data from one, cross-referencing in another, manually moving detections between platforms to build a targeting case. Maven consolidated and orchestrated all of these behind a single interface. Cameron Stanley, the Pentagon’s chief digital and AI officer, called it an “abstraction layer,” a common term in software engineering, meaning a system which hides the complexity underneath it.16 Humans run the targeting and the ML systems underneath produce confidence intervals. Three clicks convert a data point on the map into a formal detection and move it into a targeting pipeline. These targets then move through columns representing different decision-making processes and rules of engagement. The system evaluates factors and presents ranked options for which platform and munition to assign, what the military calls a Course of Action. The officer selects from the ranked options, and the system, depending on who is using it, either sends the target package to an officer for approval or moves it to execution.
The AI underneath the interface is not a language model, or at least the AI that counts is not. The systems that detect targets in satellite imagery, fuse data from radar and drone footage, and track objects across multiple intelligence sources are computer vision and sensor fusion.17 They predate large language models by years. Neither Claude nor any other LLMs detects targets, processes radar, fuses sensor data, or pairs weapons to targets. LLMs are late additions to Palantir’s ecosystem;they were added in late 2024, years after the core system was operational, “AIP” was added as a natural language layer that summarizes documents or constructs and answers queries.18 When Anthropic was blacklisted, the Pentagon signed a replacement contract with OpenAI within hours. Replacing one language model with another is often just a simple configuration change, all you really have to do is change the API endpoint.
The language model was never what mattered about this system. What mattered was what Maven did to the process: it consolidated the systems, compressed the time, and reduced the people. That is not a new idea. The United States military has been trying to close the gap between seeing something and destroying it for as long as that gap has existed, and every attempt has produced the same failure. Maven may not even be the most extreme case."
Kill Chain
On the automated bureaucratic machinery that killed 175 children
(artificialbureaucracy.substack.com)
@adfichter
thank you so much for sharing this article!! it is super useful! -
@adfichter
thank you so much for sharing this article!! it is super useful!@flaubau thank you! Everyone should read it. Cause I was wondering too: how can you use an LLM for targeting in Iran? That doesnt make sense to me.
So its is Maven=Palantir. And the kill chain term has a long history tradition in the US army.
-
"The Maven Smart System is the platform that came out of those exercises, and it, not Claude, is what is being used to produce “target packages” in Iran. There are real limits to what a civilian like myself can know about this system, and what follows is based on publicly-available information, assembled from Palantir product demos, conferences, as well as instructional material produced for military users. But we can know quite a bit. The interface looks like a tacticool, dark mode send-up of enterprise software paired with the features of geospatial application like ArcGIS. What the operator sees are either maps with GIS-like overlays or a screen organized like a project management board. There are columns representing stages of the targeting process, with individual targets moving across them from left to right, as in a Kanban board.
Before Maven, operators worked across eight or nine separate systems simultaneously, pulling data from one, cross-referencing in another, manually moving detections between platforms to build a targeting case. Maven consolidated and orchestrated all of these behind a single interface. Cameron Stanley, the Pentagon’s chief digital and AI officer, called it an “abstraction layer,” a common term in software engineering, meaning a system which hides the complexity underneath it.16 Humans run the targeting and the ML systems underneath produce confidence intervals. Three clicks convert a data point on the map into a formal detection and move it into a targeting pipeline. These targets then move through columns representing different decision-making processes and rules of engagement. The system evaluates factors and presents ranked options for which platform and munition to assign, what the military calls a Course of Action. The officer selects from the ranked options, and the system, depending on who is using it, either sends the target package to an officer for approval or moves it to execution.
The AI underneath the interface is not a language model, or at least the AI that counts is not. The systems that detect targets in satellite imagery, fuse data from radar and drone footage, and track objects across multiple intelligence sources are computer vision and sensor fusion.17 They predate large language models by years. Neither Claude nor any other LLMs detects targets, processes radar, fuses sensor data, or pairs weapons to targets. LLMs are late additions to Palantir’s ecosystem;they were added in late 2024, years after the core system was operational, “AIP” was added as a natural language layer that summarizes documents or constructs and answers queries.18 When Anthropic was blacklisted, the Pentagon signed a replacement contract with OpenAI within hours. Replacing one language model with another is often just a simple configuration change, all you really have to do is change the API endpoint.
The language model was never what mattered about this system. What mattered was what Maven did to the process: it consolidated the systems, compressed the time, and reduced the people. That is not a new idea. The United States military has been trying to close the gap between seeing something and destroying it for as long as that gap has existed, and every attempt has produced the same failure. Maven may not even be the most extreme case."
Kill Chain
On the automated bureaucratic machinery that killed 175 children
(artificialbureaucracy.substack.com)
@adfichter hat noch jemand direkt an buttle/tuttle gedacht?
-
@flaubau thank you! Everyone should read it. Cause I was wondering too: how can you use an LLM for targeting in Iran? That doesnt make sense to me.
So its is Maven=Palantir. And the kill chain term has a long history tradition in the US army.
@adfichter
AI is super old. Computer games in the 80s already used so called "Entscheidungsbäume". A couple of years before LLMs, machinelearning was a real hype in medicine because it could process so much data. It found new early cancer symptoms, it found proteins, it was a real hit.
when worries came up that AI will now be used in military I felt that LLMs are not going to be the decisionmakers. most nerds felt it would be of the more classic AI. but i figured it will have its faults because it is after all just a computer. there is no perfect computer. -
R relay@relay.infosec.exchange shared this topic
