Loading the catalogue…
Loading the catalogue…
DeepSeek is a Chinese AI lab that releases widely-used, MIT-licensed open-weights models (R1, V3.x, V4-Pro/Flash) and publishes detailed technical papers, but its hosted consumer/API service stores data in China with no EU legal entity, no Article 27 representative, and is currently subject to an Italian GDPR processing ban and a German DSA app-store takedown referral. EU regulated customers can mitigate most of this by self-hosting the open weights on EEA infrastructure, but should treat DeepSeek's own hosted API as non-compliant for personal-data workloads pending resolution of its EU representation and data-transfer mechanisms.
Italy's Garante maintains a definitive ban on DeepSeek processing Italian users' personal data, first imposed in January 2025, citing GDPR non-compliance across multiple articles.
DeepSeek's hosted consumer/API service processes and stores personal data in China, subject to PRC state-access laws, which multiple EU regulators have found unlawful under GDPR Article 44/46.
Germany's Berlin data protection commissioner referred DeepSeek to Apple and Google under DSA Article 16 for unlawful China data transfers; the Netherlands has banned DeepSeek on government devices.
A January 2025 disclosure found an unauthenticated, internet-exposed ClickHouse database leaking over a million lines of plaintext chat logs and API keys.
No EU legal entity, GDPR Article 27 representative, or public Article 28 Data Processing Agreement has been identified for DeepSeek's hosted services, complicating enforcement and contractual redress for EU customers.
DeepSeek's first external funding round (mid-2026) includes a Chinese state investment fund among its investors, deepening formal state financial ties beyond its existing High-Flyer relationship, alongside a planned mainland China IPO.
Stav’s assessment
Editorial assessment, not legal advice. Stav's risk ratings, scores, and verdicts are our own analysis of publicly available information and may be incomplete or out of date. Verify independently before making compliance or procurement decisions.
DeepSeek publishes detailed technical papers alongside major model releases, including a Nature-published R1 paper, which is unusually open compared to many closed labs.
Verified HuggingFace organisation with over 100 published models and roughly 144,000 followers; most recent flagship model weights are released under the permissive MIT licence.
DeepSeek responded promptly to Wiz Research's responsible disclosure of its exposed ClickHouse database, securing it quickly after notification.
DeepSeek continues to ship major model updates on a rapid cadence (V4-Pro general availability, new experimental multimodal models) and has attracted external institutional investment ahead of a planned public listing.
Published safeguards & certifications