- Deployment
- hosted
- Hq Country
- France
- Hq City
- Paris
- Ownership
- private
- Founded
- 2022
- Training Default
- opt-out-clear
- Residency Options
- France by default (European provider), other geographies on request, EU and US workloads on separate region-specific infrastructure
- Gov Access Exposure
- mixed
- Eu Ai Act Role
- provider
- Certs
- SOC 2 Type II
- DPA available
- yes
- Zero Retention Available
- Yes
- Retention Default
- Twelve months by default, with 1-month, 1-week, 1-day and zero-retention policies available to customers. "By default, we use a European provider based in France to respect GDPR constraints, but offer other geographies and specific providers per customer request." EU and US workloads are kept on separate region-specific infrastructure.
- Retention Exceptions
- Training is tiered by plan rather than by setting: "User audio data may be used for the training of the models. Only users in the Free Plan are subject to data used for model training. All users and customers that are in the Pro Plan or Enterprise Plan, will not have their data used for model training." The terms of use repeat it — the Free offer includes internal processing to improve the service, and free-tier output may be used to train the models. Paid tiers are excluded, so the escape is real, but it is bought rather than toggled.
- Subprocessors
- not published — the sub-processor list is available only on request through the Trust Center
- Subprocessor Count
- 0
- Transfer Mechanism
- sccs+dpf
- Eu Procurement Ready
- conditional
- Eu Procurement Reason
- A French speech-to-text vendor with a French default region, EU and US workloads on separate infrastructure, zero-retention available on request, and a real SOC 2 Type II. Two things hold it back and both are the vendor's own doing. It is the only row in this batch that does not publish its sub-processors: the compliance hub says "a maintained list of subprocessors, purposes, and locations is available" and the Trust Center provides it "upon request", so a buyer cannot see who touches the audio before signing anything. And the ISO 27001 claim does not survive reading two pages together — the body text says Gladia "is working towards ISO 27001 compliance" while the footer of every page reads "ISO 27001 Compliant". Only the SOC 2 is credited here. Free-tier audio also trains the models, which matters more than it looks: evaluation traffic is exactly what a team sends before it has a contract.
- Training Claim Basis
- stated
- Training Confidence
- high
- Residency Claim Basis
- stated
- Residency Confidence
- high
- Retention Claim Basis
- stated
- Retention Confidence
- high
- Subprocessors Claim Basis
- stated
- Subprocessors Confidence
- high
- Residency Class
- eu-default
- Portability / Exit Path
- proprietary-only
- Subprocessor Jurisdiction
- undisclosed
- Underlying model providers
- Gladia's own Solaria speech models
