Reka AI AI Vendor Risk Profile
AI research company building multimodal foundation models capable of processing text, images, video, and audio. Founded by former DeepMind and Google Brain researchers, focused on natively multimodal architectures.
Risk overview
Risk score: 34/100
Risk tier: Moderate
Safety rating: 66/100 (higher is better)
Lower risk scores indicate lower assessed risk.
Last verified: Apr 19, 2026 Aging
Risk dimensions
| Dimension | Risk score |
|---|---|
| Data Handling | 17/100 |
| IP Exposure | 25/100 |
| Jurisdiction | 8/100 |
| Security | 72/100 |
| Regulatory Compliance | 60/100 |
| Transparency | 45/100 |
| Business Stability | 64/100 |
| Dependency Chain | Not assessed |
| Agent Governance | Not assessed |
Analyst summary
Rating: Caution
Reka AI is an early-stage multimodal frontier model developer with teams across the US and Singapore. It explored an acquisition by Snowflake in 2024 that did not close, leaving the company's long-term strategic path uncertain. Compliance posture is early-stage, and enterprise buyers should treat it as a research-and-experiment vendor rather than a production standard.
Bottom line: Interesting research-grade frontier vendor; not yet a safe choice for production enterprise deployments.
Strengths
- Competitive multimodal model capability relative to team size
- Clear no-training-on-customer-data policy
- Customer retains ownership of inputs and outputs
- Published models demonstrate research credibility
Concerns
- Failed Snowflake acquisition (2024) leaves strategic future uncertain
- SOC 2 Type II described as 'pursuing' rather than completed per public posture
- Small company scale creates meaningful continuity risk for production deployments
- Limited enterprise customer references vs. OpenAI, Anthropic, Mistral
Best for
- Research teams evaluating multimodal frontier models for benchmarking
- Experimentation and prototyping where model diversity adds research value
- Startups wanting an alternative frontier vendor for pricing leverage
Avoid if
- You need a production-grade frontier model with enterprise compliance today
- Your procurement requires multi-year vendor stability commitments
- You need FedRAMP, HIPAA BAA, or ISO 27001 attestation
Citations
- Data Handling — Trains On User Data
Reka does not use customer API inputs or outputs to train its foundation models without explicit consent.
- Governance — Privacy Policy Url
Reka Privacy Policy describes data collection and use.
- Governance — Tos Url
Reka Terms of Service govern use of the API and models.
- Ip Profiles — User Owns Outputs
Customer retains ownership of Inputs and Outputs generated through the Reka API.
- Jurisdiction Profiles — Incorporation Country
Reka AI, Inc. is headquartered in San Francisco with research teams in the United States and Singapore. Reka explored acquisition by Snowflake in 2024.
- Jurisdiction Profiles — Incorporation Country
Snowflake walked away from its planned acquisition of Reka AI, reportedly valued at around $1 billion, leaving the AI startup independent.
- Security Compliance — Soc2 Type2
Reka is pursuing SOC 2 Type II compliance; enterprise deployments can run through partner clouds with SOC 2 controls.