Scale AI AI Vendor Risk Profile
Data infrastructure company providing AI training data labeling, model evaluation, and RLHF services. Operates both human-in-the-loop data annotation at scale and AI-powered evaluation tools used by leading AI labs and enterprises.
Risk overview
Risk score: 23/100
Risk tier: Moderate
Safety rating: 77/100 (higher is better)
Lower risk scores indicate lower assessed risk.
Last verified: Apr 19, 2026 Aging
Risk dimensions
| Dimension | Risk score |
|---|---|
| Data Handling | 20/100 |
| IP Exposure | 21/100 |
| Jurisdiction | 13/100 |
| Security | 22/100 |
| Regulatory Compliance | 25/100 |
| Transparency | 50/100 |
| Business Stability | 37/100 |
| Dependency Chain | 23/100 |
| Agent Governance | Not assessed |
Analyst summary
Rating: Recommended
Scale AI is the dominant data-labeling and model-evaluation vendor, with deep US defense and intelligence-community engagement (Scale Donovan for classified environments, Thunderforge DoD contract). Meta's June 2025 $14.3 billion acquisition of a 49 percent stake and the hiring of founder Alexandr Wang raised antitrust concerns and created operational uncertainty for non-Meta customers.
Bottom line: Technically strong and mature for defense and enterprise workloads; the Meta stake reshuffles the competitive risk profile.
Strengths
- No training on customer data; customer datasets isolated per engagement
- SOC 2 Type II, ISO 27001 certified with FedRAMP Moderate authorization
- On-premises and air-gapped deployments available (Scale Donovan for classified workloads)
- Deep DoD and intelligence community engagements signal mature high-assurance delivery
- Customer retains full ownership of labeled datasets and model evaluations
Concerns
- Meta's 49 percent stake (June 2025, $14.3B) creates conflict-of-interest and data-leakage concerns for Meta's competitors
- Alexandr Wang departed to lead Meta's superintelligence efforts, removing founder focus from Scale
- FTC scrutiny of the Meta deal adds ongoing regulatory uncertainty
- Contractor workforce model creates supply-chain and labor-ethics exposure
Best for
- US defense and intelligence community workloads needing classified-environment data labeling
- Enterprises building custom models at scale needing high-assurance data labeling
- Model evaluation and red-teaming programs requiring mature evaluator workforces
Avoid if
- You are a direct competitor to Meta (Meta's 49 percent stake creates clear conflicts)
- You cannot tolerate the strategic uncertainty post-Wang departure
- Your engagement requires founder-level continuity
Citations
- Data Handling — Data Residency Options
Scale offers on-premises and air-gapped deployments for government and defense customers, including Scale Donovan for classified environments.
- Data Handling — Third Party Data Sharing
Scale AI uses a vetted network of contractors to label data; access is controlled by tiered NDAs and role-based permissions.
- Data Handling — Trains On User Data
Scale AI does not use customer-submitted data to train foundation models. Customer datasets are isolated per engagement.
- Governance — Government Contracts
Scale AI was awarded a Thunderforge contract by the Department of Defense for AI-enabled military planning, and holds significant US defense and intelligence community engagements.
- Governance — Strategic Investors
Meta acquired a 49 percent stake in Scale AI in a $14.3 billion investment and hired founder Alexandr Wang to lead its superintelligence efforts.
- Ip Profiles — User Owns Outputs
Customer retains all right, title, and interest in Customer Data and labeled datasets delivered by Scale.
- Jurisdiction Profiles — Incorporation Country
Scale AI, Inc. is a Delaware corporation headquartered in San Francisco, California.
- Security Compliance — Iso 27001
Scale AI holds ISO/IEC 27001 certification and FedRAMP Moderate authorization for federal workloads.
- Security Compliance — Soc2 Type2
Scale AI maintains SOC 2 Type II certification covering its platform and data labeling services.