DeepSeek via Fireworks AI

Fireworks' security docs state the platform is SOC 2 Type II compliant and that "documentation and audit reports are available in our Trust Center" (trust.fireworks.ai). The report itself is gated behind Trust Center access; the attestation of SOC 2 Type II status is public. This covers the Fireworks serving platform, not DeepSeek the developer.

tier: self_serve · route: trust_center_nda ·

source · archived copy

Fireworks docs state ISO 27001 certification (plus ISO 27701 privacy extension) with certificate PDFs downloadable from the Trust Center. Platform-level certification (Fireworks, not DeepSeek).

tier: self_serve · route: trust_center_nda ·

also_certified: ['ISO 27701']

source · archived copy

Fireworks docs list ISO 42001 (AI management system) among its certifications, with certificate PDFs available via the Trust Center. Certifies Fireworks' AI management practices as the serving platform; it says nothing about DeepSeek's model development practices.

tier: self_serve · route: trust_center_nda ·

source · archived copy

Maintained SafeBase-style trust center; hosts certifications, audit reports, and (per the DPA) subprocessor-change notifications customers can subscribe to. Page is JS-rendered; no Wayback snapshot exists as of verification date.

tier: self_serve · route: public ·

source

Fireworks publicly states the inference platform is HIPAA compliant (docs and enterprise page), which conventionally implies BAA availability, but no public BAA terms were found and the public DPA contains no BAA or HIPAA language. Treat BAA execution as a sales conversation; confirm whether serverless DeepSeek endpoints are in BAA scope or dedicated deployments are required.

tier: enterprise_only · route: sales_contract · default: requires_approval ·

source · archived copy

Public DPA (PDF) incorporates the EU Standard Contractual Clauses (Commission Implementing Decision (EU) 2021/914, Module Two) plus Swiss addendum provisions, and publishes the authorised sub-processor list as Schedule 4 (cloud infrastructure: AWS in the US and Japan, GCP, Oracle Cloud). Subprocessor changes are notified via the Trust Center. DeepSeek is not a sub-processor. Privacy policy separately confirms SCCs for UK/EEA transfers.

tier: self_serve · route: public ·

sccs: EU 2021/914 Module Two · subprocessors_listed: ['Amazon Web Services', 'Google Cloud Platform', 'Oracle Cloud']

source

Privacy policy: "We do not use your prompts, training data, or API inputs to train or improve our AI models without your explicit opt-in." The public DPA reinforces this contractually, prohibiting "using Covered Data to train, fine-tune, or otherwise improve any shared or foundational model." Because Fireworks hosts open weights, customer data also cannot reach DeepSeek for training.

tier: self_serve · route: public · default: enabled ·

source · archived copy

Zero data retention is the documented default for open models: prompt and generation data "exist only in volatile memory for the duration of the request" and are not written to persistent storage without explicit opt-in. The public DPA makes this contractual when the service is used in its Zero Data Retention configuration. Exception: the Responses API stores conversation state for 30 days when store=true (that API's default), set store=false for strict ZDR. Confirm your integration path stays within the ZDR configuration.

tier: self_serve · route: public · default: enabled ·

responses_api_retention_days: 30

source · archived copy

The core point of this row: Fireworks, a US-based company, serves open-weight DeepSeek models entirely on its own infrastructure (AWS, GCP, Oracle Cloud data centers per its DPA, US, Japan, and EU locations). There is no data path to DeepSeek the company: DeepSeek is not a subprocessor and inference does not call DeepSeek's API; the "China data path" of DeepSeek's first-party service does not apply here. Region pinning (including EUROPE: Frankfurt and Iceland) is available for on-demand/dedicated deployments; the region of default serverless inference is not publicly specified, and DPA clause 6.1 permits processing anywhere Fireworks or its sub-processors maintain facilities, hence partial. Enterprise page claims "full support for data residency" via sales.

tier: self_serve · route: public · default: requires_config · geography: EU region available (Frankfurt, Iceland) for dedicated deployments

eu_locations: ['Frankfurt', 'Iceland'] · regions_multi: ['GLOBAL', 'US', 'EUROPE', 'APAC']

source

Per the two-level rule this dimension describes DeepSeek, the model developer, the GPAI Code of Practice is a provider obligation, and for open-weight models the GPAI provider remains DeepSeek, not Fireworks. DeepSeek does not appear on the EC's signatory list (23 signatories as of 2026-07-05; xAI is a partial Safety & Security-only signatory). Absence is a finding, though the Code is voluntary and providers may demonstrate AI Act compliance by alternative means. Nuance: some Art 53 obligations are lightened for open-weight releases meeting Art 53(2) conditions, but signatory status is simply absent here.

signatory_count_at_check: 23

source · archived copy

Describes DeepSeek as the GPAI provider (not Fireworks). No public training-content summary using the EC's mandatory template (published 24 July 2025) was found for DeepSeek V3/R1. Context: for models placed on the market before 2 August 2025, which covers the V3/R1 family, the EC allows summaries to be published as late as 2 August 2027, so absence today is a transparency gap rather than a proven violation. DeepSeek publishes technical reports on training methodology, but these do not follow the Art 53(1)(d) template.

no public source