Is Cohere safe with your data?

Cohere

55/100

Partial disclosure · high confidence

Cohere earns a C (55/100) because it discloses its data practices only in part.

Dealbreaker flag

D1.1: For Trial Users and Researchers...we may use this input and output data to conduct research and development
D1.4: YOU GRANT US A NONEXCLUSIVE, WORLDWIDE, ROYALTY-FREE, IRREVOCABLE, SUBLICENSABLE...RIGHT...including by sharing API Data and Finetuning Data with Third Parties

#66

of 211 apps ranked

score · Model provider avg 57

-2

vs category average

Grade scaleA · 70–100B · 60–69C · 48–59D · 35–47F · 0–34

Cohere trains trial users' inputs on models with no opt-out, offers vague data controls to enterprise customers, and grants itself irrevocable sublicensable rights to share training data with unlimited third parties.

What Cohere's privacy policy and terms of service say about your data

D1.4: Perpetual sublicensable licence

Cohere explicitly grants itself 'IRREVOCABLE, SUBLICENSABLE' rights and reserves the right to share training and fine-tuning data with third parties 'in accordance with their terms and conditions' , with no user opt-out.

D1.1: Trial users' data trained with no opt-out

For Trial Users and Researchers, policy states 'We may use this input and output data to conduct research and development' with only a de-identification safeguard, no opt-out mechanism disclosed.

D3.1: Enterprise-only 30-day retention

Only Enterprise Users get a defined 30-day retention period. Trial Users and Researchers receive no named retention commitment.

D2.1–D2.4: Conditional deletion/access rights

All data subject rights (access, deletion, correction, portability) are qualified with 'may request', 'subject to certain exceptions', and caveats , no unconditional commitments.

What the policy is silent or vague on

Not stated: keeping user inputs out of model training
Not stated: your ownership of generated outputs
Not stated: a deletion timeline after closure or request
Not stated: shorter retention for AI conversation logs

Cohere privacy rating

Training-data use1 of 4 disclosed

Keeps user inputs out of model training, or makes training opt-inAdverse

Names a way to opt out of or into trainingPartial

Says whether training use differs by plan or tierDisclosed

Lets the user keep ownership of generated outputsAdverse

Data-subject rights1 of 5 disclosed

Grants a right to access your dataPartial

Grants a right to delete your dataPartial

Offers data portability in a usable formatPartial

Grants a right to correct your dataPartial

Grants a way to object to or opt out of processingDisclosed

Retention and deletion2 of 4 disclosed

States a retention period for your dataDisclosed

States a deletion timeline after closure or requestSilent

Sets a shorter retention for AI conversation logsSilent

Commits to collecting only the data it needsDisclosed

Third-party sharing3 of 5 disclosed

Lists the categories of third parties it shares withDisclosed

References a sub-processor list or data processing agreementDisclosed

Does not sell or share data for advertising, or offers opt-outSilent

Names a safeguard for international data transfersDisclosed

States a standard for government and law-enforcement accessPartial

Transparency2 of 4 disclosed

Discloses that you are interacting with AIPartial

Marks AI-generated or synthetic outputNot applicable

Enumerates the categories of data it collectsDisclosed

Maps processing purposes to legal basesDisclosed

Is versioned and dated, with change noticePartial

Sensitive data and children1 of 2 disclosed

Discloses automated decisions and a human-review pathNot applicable

Limits the use of special-category dataSilent

Governs biometric data specificallyNot applicable

States protections for children's dataDisclosed

Security and accountability0 of 3 disclosed

Describes its security safeguardsPartial

Commits to breach notificationSilent

Names a certification or a privacy contactPartial

DisclosedPartialSilentAdverseNot applicable

Details

Category: Model provider
Modalities: text
Processes biometrics: No
Policy last updated: 2026-05-01
Region scored: Global / US-default
Last assessed: 2026-07-08

Documents examined

Other model provider apps

Fireworks AIC

Abacus.AID

Together AID

Each grade reflects our analysis of what an app states in its public privacy policy and terms as of the assessment date. It measures the transparency of those documents, not the company's actual data practices, security, or compliance. Grades are our opinion, offered for general information. Full disclaimer.