◉ PSYCHOHISTORY

Scale AI (2016) — DoD Data-Labeling Substrate (Project Maven, Thunderforge, JADC2)

corporationAI & Compute
A single company labels the data that trains both the Pentagon's AI and the biggest chatbots — a quiet chokepoint over what AI learns.
Who they are

Scale AI, founded in 2016 by Alex Wang, a data-labeling company for machine learning.

What they do

The engine calls it an indispensable chokepoint that formats real-world data into the training sets that shape AI.

How it works

Its customers include the Department of Defense (Project Maven computer vision, plus Thunderforge and JADC2 contracts, verified via USASpending.gov) and frontier AI labs like OpenAI and Anthropic; by labeling both military and commercial data it gives the intelligence community structural visibility into the data that shapes AI model behavior.

Why it matters

The engine holds several readings at once — a deliberate sole-source chokepoint, an ordinary case of the military outsourcing to specialists, and a plain commercial service — all seen as true simultaneously. Whoever shapes the training data shapes the AI.

The engine's record — word for word
Founded 2016 by Alex Wang in San Francisco. Function: large-scale human + AI-assisted data labeling and annotation pipelines for ML training. Major customers: Department of Defense (Project Maven computer vision, Thunderforge, JADC2 component contracts), OpenAI / Anthropic (training-data services), other frontier-model labs. Primary-source DoD contracts: USASpending.gov FA714622CB007 (2022, DoD) + SP470125F0273 (2025, DoD recompete). Engine relevance: the indispensable data-labeling chokepoint that formats real-world military + commercial data into machine-readable training sets, granting the IC structural visibility into (and influence over) the data substrate that shapes frontier-model weights. Per Report #93 H2 findings #011 + #012 — primary-source verified via USASpending.gov. Apex (a) coordinated-sole-source-chokepoint + (b) structural-recurrence of military-outsourcing-to-civilian-specialists + (c) compound-null commercial-B2B-services all load-bearing simultaneously.
Follow the trail
Walk this on the live map →
Part of the Psychohistory engine — 2,437 entities, 6,337 documented connections. Open data, built to be proven wrong.