Scale AI

Scale AI is a data-labeling, human-feedback, and AI evaluation company founded in 2016 by Alexandr Wang (then 19, an MIT dropout). It became the dominant supplier of labeled data and reinforcement-learning-from-human-feedback (RLHF) labor to the frontier AI industry, supplying training pipelines at OpenAI, Anthropic, Google, Meta, and Microsoft, and has since expanded into government AI integration and model evaluation. In June 2025, Meta took a roughly 49% non-voting stake in Scale for $14.3 billion, with Wang departing to become Meta's Chief AI Officer.

Snapshot

Government contracts

Both contracts were disclosed by interim CEO Jason Droege in April 2026 and were signed after Meta took its 49% / $14.3B stake.

Business lines

Scale's historical core is data labeling and human feedback for frontier labs, drawing on a global contractor workforce operated under the Remotasks and Outlier brands. It has since added the Scale GenAI Platform for enterprise fine-tuning and deployment, the SEAL evaluation business, a federal-government integration arm (Scale AI Federal, including the Scale AI Defense LLC subsidiary), and Donovan, an LLM-powered decision-support product for defense and intelligence users.

SEAL evaluations

Scale's SEAL (Safety, Evaluations, and Alignment Lab) runs private holdout leaderboards for coding, math, instruction-following, adversarial robustness, and multilinguality. Because the evaluation questions are not released, SEAL scores are considered less contaminatable than public benchmarks. SEAL is one of the few private-sector model-evaluation operations with industry credibility, alongside Epoch AI and the government-run UK AISI and US AISI bodies. SEAL's private holdouts are cited in policy debates over whether third-party evaluations can substitute for government testing, including discussions around America's AI Action Plan and California SB 53 — Transparency in Frontier AI Act.

Government AI integration

Scale has become a principal integrator of AI for the US national-security customer, and is, along with Palantir, Anduril, and Microsoft, one of a small number of firms delivering operational AI capability to the Department of Defense and intelligence community. Its principal programs include:

• Thunderforge (Feb 2025) — Selected by DIU/CDAO as the prime integrator for a DoD program bringing generative AI to joint operational planning at EUCOM and INDOPACOM, partnered with Microsoft and Anduril.
• Army TITAN — Deep-sensing targeting platform; Scale provides AI/ML components.
• Scale AI Defense LLC — Dedicated defense subsidiary with its own security clearances and contracting vehicle.
• State and federal civilian — Growing contracts across the VA, Treasury, and state digital-services offices.

In an April 2026 interview, Droege said the United States had not yet had its AI "Sputnik moment" but predicted one within the next year, and disclosed two contracts signed since the Meta investment: a deal with the Pentagon's CDAO carrying a $100 million ceiling and a $99.5 million U.S. Army contract (Source: politico.com). Customers' treatment of Meta's stake as a conflict of interest has shifted Scale's revenue mix toward government work.

Key people

• Alexandr Wang (founder; Meta Chief AI Officer since June 2025): Led Meta's new "Superintelligence Labs" unit on departure. Prior to leaving, he was among the most-quoted industry voices on US-China AI competition and testified frequently to Congress, including on US export controls.
• Lucy Guo (co-founder; departed 2018): Later founded Passes.
• Jason Droege (acting CEO since June 2025): Former Uber Eats founder; previously Scale COO.
• Vijaye Raji (CTO): Former Meta engineering leadership.

Meta transaction (June 2025)

In June 2025, Meta invested $14.3 billion for a roughly 49% non-voting stake in Scale. Wang and about 50 senior Scale staff joined Meta directly, with Wang becoming Chief AI Officer and leading Meta's "Superintelligence Labs" unit. Scale continued to operate as an independent company under Droege as interim CEO. The deal was structured to avoid formal merger-review thresholds, mirroring the structural template of Microsoft-Inflection (March 2024) and Google-Character.AI (August 2024), an "acqui-hire without merger filing" pattern. It is under FTC review and was the subject of congressional letters in late 2025, and it figured in the successor concerns of the FTC's January 2025 FTC 6(b) Staff Report: Partnerships Between Cloud Service Providers and AI Developers report.

Several frontier-lab customers, including OpenAI, Google DeepMind, and Anthropic, reduced or paused Scale work after the deal, citing concerns about exposing competitively sensitive data to a company in which Meta holds a stake.

Controversies

• Contractor labor practices: Sustained criticism over Remotasks/Outlier contractor pay, working conditions, and the psychological impact of content-moderation labeling, with class-action suits in California and the Philippines.
• Defense work: Scale's explicit government defense posture has drawn internal and activist pushback, though less intensely than at OpenAI or Google.
• Customer conflict post-Meta: Frontier-lab customers treating Meta's stake as a conflict-of-interest event has reshaped Scale's revenue mix toward government.
• Data integrity: Recurring questions about whether training labels provided to competing labs remain adequately siloed.

Relevance to AI policy

Scale sits at a concentration point in the AI training supply chain, supplying labeled data and human feedback; concerns about that chokepoint (concentration, labor standards, quality) cut across AI governance regimes. The Meta-Scale deal is a reference point in debates over whether AI M&A rules need updating, directly relevant to FTC 6(b) Staff Report: Partnerships Between Cloud Service Providers and AI Developers follow-ons. Through its defense programs Scale is one of a handful of firms delivering operational AI capability to the DoD and intelligence community, alongside Palantir, Anduril, and Microsoft. SEAL's private holdouts feature in debates over third-party versus government evaluation. On export controls, Wang was a prominent public voice, and Scale is a downstream beneficiary of preserved US compute leadership per Export Controls (AI).

Relationships

• related: Meta AI — principal investor (~49%, June 2025); absorbed Wang and senior leadership
• related: OpenAI, Anthropic, Google DeepMind — historic customers; reduced relationships post-Meta deal
• related: Palantir Technologies, Anduril Industries — peer government-AI integrators
• related: Chief Digital and Artificial Intelligence Office (CDAO) — DoD customer; Thunderforge prime-integrator relationship
• related: FTC 6(b) Staff Report: Partnerships Between Cloud Service Providers and AI Developers — Meta-Scale deal is the next-generation version of the partnerships the FTC examined
• related: AI and National Security
• related: Export Controls (AI)
• related: Clawed — part of the government-AI integrator landscape around the Anthropic-DoW dispute

See also

• Clawed — context for the government-AI integrator landscape
• Jason Droege — interim CEO