SCALE AI

Institutional Archetype

THE REFEREE THAT A TEAM BOUGHT HALF OF — Scale AI labels the training data the labs learn from, co-wrote one of the hardest public exams they are graded on, and in February 2025 became the US government’s authorized outside evaluator of AI models. Four months later Meta bought 49 percent of it and hired its founder to run Meta’s superintelligence lab. Each step was announced. The file is open because the same company now supplies the labs, grades them for the government, and is part-owned by one of them.

Mandate & Origin

A data company for AI labs, enterprises and governments: “Scale will continue to partner with leading AI labs, multinational enterprises, and governments to deliver expert data and technology solutions through every phase of AI’s evolution” (Scale AI, June 12 2025). Its research arm is SEAL.

Funding & Backers

Meta, June 2025. Scale announced “a significant new investment from Meta Platforms” that “values Scale at over $29 billion”; “Following its investment, Meta will hold a minority of Scale’s outstanding equity. Scale remains an independent leader in AI” (Scale AI). The Associated Press reported the investment as $14.3 billion for a 49 percent stake (AP via WJXT).

Position in the Apparatus

• The government’s evaluator. On February 10, 2025 Scale announced that it and the US AI Safety Institute (now CAISI) were “partnering to develop improved methods to test frontier AI models,” with “novel evaluations jointly developed by the AISI and Scale’s research arm”; model builders could “test their models once and share results with a global network of AI safety institutes, if desired” (Scale AI). FedScoop reported it as the first third-party evaluator the institute authorized (FedScoop).
• The exam. Humanity’s Last Exam was built by the Center for AI Safety and Scale; its senior authors are Dan Hendrycks, Alexandr Wang and Summer Yue (arXiv 2501.14249). Hendrycks also advises Scale.
• The founder’s move. “Scale’s founder, Alexandr Wang, is joining Meta to work on Meta’s AI efforts. Wang will continue to serve as a director on the Scale Board of Directors”; Chief Strategy Officer Jason Droege became interim CEO (Scale AI). On June 30, 2025 Zuckerberg’s memo announced that Wang “has joined Meta to serve as our Chief AI Officer and lead MSL” (CNBC via NBC New York).
• The other labs’ answer. In July 2025 Scale laid off about 200 employees, 14 percent of its workforce, and 500 contractors; TechRepublic noted reports that “companies such as OpenAI and Google are terminating their partnerships with Scale AI,” and that OpenAI “denied that the decision was related to any risk of Meta accessing its trade secrets” (TechRepublic).

Actions & Leadership Choices

CONDUCT: CONFLICTED — GOVERNMENT EVALUATOR AND LAB SUPPLIER, PARTLY OWNED BY A LAB. No wrongdoing is alleged. The record shows the market’s own verdict on independence: when one lab bought half of the company, competing labs reportedly took their data work elsewhere. The government evaluation arrangement announced in February 2025 has no public statement in the sources above on how it is handled now that a lab holds 49 percent and the founder runs that lab’s AI. That is the question the file holds open.

Reach Assessment

Institutional: Training data for the labs, a government evaluation agreement, and a benchmark the field cites. Memetic: “Humanity’s Last Exam” is the name the press uses for the frontier.

Sources: Scale AI Announces Next Phase of Company’s Evolution — Scale AI (June 12, 2025); First independent model evaluator for the USAISI — Scale AI (Feb 10, 2025); US AI Safety Institute taps Scale AI for model evaluation — FedScoop; Meta invests in AI firm Scale and recruits its CEO — AP via WJXT; Zuckerberg’s memo announcing Meta Superintelligence Labs — CNBC via NBC New York; Humanity’s Last Exam — arXiv 2501.14249; Scale AI Lays Off Hundreds After $14.3B Meta Investment — TechRepublic.