DeepSeek chat adds image-to-text, raising privacy concerns over data handling in China

DeepSeek has introduced an image recognition feature in its chat application, powered by the DeepSeek-VL2 architecture, allowing users to extract text from images like receipts and handwritten notes. However, concerns have been raised regarding the privacy implications of uploading sensitive documents, as data processed by DeepSeek is stored in China and may be shared with government authorities. This has led to scrutiny, with evidence suggesting potential data transmission to infrastructure linked to China Mobile. Separately, DeepSeek-OCR and DeepSeek-OCR2 are available as open-source models for local processing, offering an alternative for users prioritizing data privacy. AI

IMPACT Raises privacy concerns for users uploading sensitive documents to cloud-based AI services, highlighting the trade-off between convenience and data security.

RANK_REASON The article discusses a new feature in an existing product (DeepSeek chat's image recognition) and its implications, rather than a novel model release or fundamental research breakthrough.

Read on dev.to — LLM tag →

• China Mobile
• CUDA
• DeepSeek
• DeepSeek-OCR
• DeepSeek-OCR2
• DeepSeek-VL2
• Feroot Security
• GigaChat
• IT Home
• Ivan Tsarynny
• PyTorch

• product
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →