DeepSeek (chatbot)

On 10 January 2025, DeepSeek released the chatbot, based on the DeepSeek-R1 model, for iOS and Android. By 27 January, DeepSeek-R1 surpassed ChatGPT as the most-downloaded freeware app on the iOS App Store in the United States, which resulted in an 18% drop in Nvidia's share price.{{Cite web |last=Field|first=Hayden|date=27 January 2025|title=China's DeepSeek AI dethrones ChatGPT on App Store: Here's what you should know |url=https://www.cnbc.com/2025/01/27/chinas-deepseek-ai-tops-chatgpt-app-store-what-you-should-know.html|work=CNBC|access-date=4 February 2025|archive-date=28 January 2025|archive-url=https://web.archive.org/web/20250128201052/https://www.cnbc.com/2025/01/27/chinas-deepseek-ai-tops-chatgpt-app-store-what-you-should-know.html|url-status=live}}{{Cite web |last=Picchi|first=Aimee|date=27 January 2025|title=What is DeepSeek, and why is it causing Nvidia and other stocks to slump?|url=https://www.cbsnews.com/news/what-is-deepseek-ai-china-stock-nvidia-nvda-asml/|work=CBS News|access-date=4 February 2025 |archive-date=29 January 2025|archive-url=https://web.archive.org/web/20250129002522/https://www.cbsnews.com/news/what-is-deepseek-ai-china-stock-nvidia-nvda-asml/|url-status=live}} After a "large-scale" cyberattack on the same day disrupted the proper functioning of its servers, DeepSeek limited its new user registration to phone numbers from mainland China, email addresses, or Google account logins.{{cite news|last=Kerr |first=Dara|date=27 January 2025|title=DeepSeek hit with 'large-scale' cyber-attack after AI chatbot tops app stores |url=https://www.theguardian.com/technology/2025/jan/27/deepseek-cyberattack-ai|access-date=28 January 2025|work=The Guardian}}{{Cite web|last1=Tweedie|first1=Steven|last2=Altchek |first2=Ana|date=27 January 2025|title=DeepSeek temporarily limited new sign-ups, citing 'large-scale malicious attacks' |url=https://www.businessinsider.com/deepseek-issues-account-registration-limited-china-phone-numbers-malicious-attack-2025-1|url-status=live|archive-url=https://web.archive.org/web/20250202052738/https://www.businessinsider.com/deepseek-issues-account-registration-limited-china-phone-numbers-malicious-attack-2025-1|archive-date=2 February 2025|access-date=4 February 2025|website=Business Insider}}

On 3 April 2025, in collaboration with researchers at Tsinghua University, DeepSeek published a paper unveiling a new model that combines the techniques generative reward modeling (GRM) and self-principled critique tuning (SPCT). The resulting model is referred to as DeepSeek-GRM. The goal of using these techniques is to foster more effective inference-time scaling within their LLM and chatbot services. Notably DeepSeek has said that these new models will be released and made open source.{{cite news |last1=Shen |first1=Xinmei |title=DeepSeek unveils new AI reasoning method as anticipation for its next-gen model rises |url=https://www.scmp.com/tech/tech-trends/article/3305259/deepseek-unveils-new-ai-reasoning-method-anticipation-its-next-gen-model-rises?module=perpetual_scroll_0&pgtype=article |access-date=6 April 2025 |work=South China Morning Post |date=6 April 2025}}{{cite arXiv |last1=Liu |first1=Zijun |last2=Wang |first2=Peiyi |last3=Xu |first3=Runxin |last4=Ma |first4=Shirong |last5=Ruan |first5=Chong |last6=Li |first6=Peng |last7=Liu |first7=Yang |last8=Wu |first8=Yu |title=Inference-Time Scaling for Generalist Reward Modeling |date=3 April 2025 |class=cs.CL |eprint=2504.02495}}

On 30 April 2025, Deepseek released its math-focused Artificial Intelligence Model named "DeepSeek-Prover-V2-671B".{{Cite web |last=Wiggers |first=Kyle |date=2025-04-30 |title=DeepSeek upgrades its math-focused AI model Prover |url=https://techcrunch.com/2025/04/30/deepseek-upgrades-its-ai-model-for-math-problem-solving/ |access-date=2025-05-06 |website=TechCrunch}} This model is useful for formal theorem proving and mathematical reasoning.{{Cite web |last=Wiggers |first=Kyle |date=2025-04-30 |title=DeepSeek upgrades its math-focused AI model Prover |url=https://techcrunch.com/2025/04/30/deepseek-upgrades-its-ai-model-for-math-problem-solving/ |access-date=2025-04-30 |website=TechCrunch}}{{Cite web |date=2025-04-30 |title=DeepSeek quietly updates open-source model that handles maths proofs |url=https://www.scmp.com/tech/tech-trends/article/3308566/deepseek-quietly-updates-open-source-model-handles-maths-proofs |access-date=2025-04-30 |website=South China Morning Post}}

DeepSeek can answer questions, solve logic problems, and write computer programs on par with other chatbots, according to benchmark tests used by American AI companies.

Users can access the chatbot for free through the official DeepSeek website or mobile application, without limitation on the number of queries.{{cite news|title=DeepSeek or ChatGPT? A price-to-performance comparison. What you need to know |url=https://economictimes.indiatimes.com/magazines/panache/deepseek-or-chatgpt-a-price-to-performance-comparison-what-you-need-to-know/articleshow/117636306.cms|access-date=4 April 2025|work=The Economic Times}} DeepSeek only supports user-signup via a global email service, e.g. Gmail, Google or Yahoo. DeepSeek also offers access to the R1 and V3 models that power the chatbot via an API with a usage-based pricing model.{{cite web |title=API Pricing |url=https://api-docs.deepseek.com/quick_start/pricing |website=DeepSeek API Documentation |access-date=4 April 2025}} This modality is primarily targeted towards developers and businesses. As of February 2025, API usage is priced at approximately $0.55 per million input tokens and $2.19 per million output tokens, making it less expensive than some competing services.

DeepSeek-V3 uses significantly fewer resources compared to its peers. For example, whereas the world's leading AI companies train their chatbots with supercomputers using as many as 16,000 graphics processing units (GPUs), DeepSeek claims to have needed only about 2,000 GPUs—namely, the H800 series chips from Nvidia.{{cite arXiv|eprint=2412.19437 |author1=DeepSeek-AI |last2=Liu |first2=Aixin |last3=Feng |first3=Bei |last4=Xue |first4=Bing |last5=Wang |first5=Bingxuan |last6=Wu |first6=Bochao |last7=Lu |first7=Chengda |last8=Zhao |first8=Chenggang |last9=Deng |first9=Chengqi |last10=Zhang |first10=Chenyu |last11=Ruan |first11=Chong |last12=Dai |first12=Damai |last13=Guo |first13=Daya |last14=Yang |first14=Dejian |last15=Chen |first15=Deli |last16=Ji |first16=Dongjie |last17=Li |first17=Erhang |last18=Lin |first18=Fangyun |last19=Dai |first19=Fucong |last20=Luo |first20=Fuli |last21=Hao |first21=Guangbo |last22=Chen |first22=Guanting |last23=Li |first23=Guowei |last24=Zhang |first24=H. |last25=Bao |first25=Han |last26=Xu |first26=Hanwei |last27=Wang |first27=Haocheng |last28=Zhang |first28=Haowei |last29=Ding |first29=Honghui |last30=Xin |first30=Huajian |title=DeepSeek-V3 Technical Report |date=2024 |class=cs.CL |display-authors=1 }}{{Cite web |last=Cosgrove |first=Kelsey Vlamis, Emma |title=Nvidia probed over how its chips may have been obtained by DeepSeek, which US lawmakers accused of spying for China |url=https://www.businessinsider.com/lawmakers-deepseek-spying-china-nvidia-probed-chips-2025-4 |access-date=2025-05-06 |website=Business Insider}} It was trained in around 55 days at a cost of US$5.58 million, which is roughly one-tenth of what tech giant Meta spent building its latest AI technology.

DeepSeek's success against larger and more established rivals has been described as "upending AI", constituting "the first shot at what is emerging as a global AI space race", and ushering in "a new era of AI brinkmanship".{{Cite news|last=Roose |first=Kevin|date=28 January 2025|title=Why DeepSeek Could Change What Silicon Valley Believe About A.I. |url=https://www.nytimes.com/2025/01/28/technology/why-deepseek-could-change-what-silicon-valley-believes-about-ai.html|access-date=28 January 2025|work=The New York Times|issn=0362-4331}}

DeepSeek's competitive performance at relatively minimal cost has been recognized as potentially challenging the global dominance of American AI models.{{cite news|date=27 January 2025|title=Chinese AI startup DeepSeek overtakes ChatGPT on Apple App Store|url=https://www.reuters.com/technology/artificial-intelligence/chinese-ai-startup-deepseek-overtakes-chatgpt-apple-app-store-2025-01-27/|access-date=27 January 2025|website=Reuters}} Various publications and news media, such as The Hill and The Guardian, have described the release of the R1 chatbot as a "Sputnik moment" for American AI,{{Cite news|last=Wade|first=David|date=6 December 2024 |title=American AI has reached its Sputnik moment |url=https://thehill.com/opinion/technology/5024271-ai-sputnik-moment-china-challenge/|url-status=live|archive-url=https://web.archive.org/web/20241208215538/https://thehill.com/opinion/technology/5024271-ai-sputnik-moment-china-challenge/|archive-date=8 December 2024|access-date=25 January 2025|work=The Hill}}{{Cite web|last1=Milmo|first1=Dan|last2=Hawkins|first2=Amy |last3=Booth|first3=Robert|last4=Kollewe|first4=Julia|date=28 January 2025|title='Sputnik moment': $1tn wiped off US stocks after Chinese firm unveils AI chatbot |url=https://www.theguardian.com/business/2025/jan/27/tech-shares-asia-europe-fall-china-ai-deepseek|work=The Guardian}}{{Cite web|last1=Hoskins|first1=Peter|last2=Rahman-Jones|first2=Imran|date=27 January 2025|title=Nvidia shares sink as Chinese AI app spooks markets |url=https://www.bbc.com/news/articles/c0qw7z2v1pgo|url-status=live |archive-url=https://web.archive.org/web/20250128001127/https://www.bbc.com/news/articles/c0qw7z2v1pgo|archive-date=28 January 2025|access-date=28 January 2025|work=BBC}} echoing Marc Andreessen's view. OpenAI wrote a letter to the Office of Science and Technology Policy (OSTP), in March 2025, citing issues concerning a possibility that Deepseek could manipulate responses to cause harm.{{Cite news |last1=Tobin |first1=Meaghan |last2=Fu |first2=Claire |date=2025-03-18 |title=From Courtrooms to Crisis Lines, Chinese Officials Embrace DeepSeek |url=https://www.nytimes.com/2025/03/18/business/china-government-deepseek.html |access-date=2025-04-10 |work=The New York Times |issn=0362-4331}}

DeepSeek's founder Liang Wenfeng has been compared to OpenAI CEO Sam Altman, with CNN calling him the Sam Altman of China and an evangelist for AI.{{Cite web|last=Goldman|first=David |date=27 January 2025|title=What is DeepSeek, the Chinese AI startup that shook the tech world? {{!}} CNN Business |url=https://www.cnn.com/2025/01/27/tech/deepseek-ai-explainer/index.html|access-date=29 January 2025|work=CNN|archive-date=29 January 2025|archive-url=https://web.archive.org/web/20250129183147/https://www.cnn.com/2025/01/27/tech/deepseek-ai-explainer/index.html|url-status=live}} Chinese state media widely praised DeepSeek as a national asset.{{Cite news|date=29 January 2025|title=DeepSeek poses a challenge to Beijing as much as to Silicon Valley |url=https://www.economist.com/business/2025/01/29/deepseek-poses-a-challenge-to-beijing-as-much-as-to-silicon-valley|url-access=subscription|access-date=31 January 2025|newspaper=The Economist|issn=0013-0613|archive-date=31 January 2025|archive-url=https://web.archive.org/web/20250131042816/https://www.economist.com/business/2025/01/29/deepseek-poses-a-challenge-to-beijing-as-much-as-to-silicon-valley|url-status=live}}{{Cite news |last1=Paul|first1=Katie|last2=Nellis|first2=Stephen|date=30 January 2025|title=Chinese state-linked accounts hyped DeepSeek AI launch ahead of US stock rout, Graphika says |url=https://www.reuters.com/technology/artificial-intelligence/chinese-state-linked-accounts-hyped-deepseek-ai-launch-ahead-us-stock-rout-2025-01-31/|access-date=30 January 2025|website=Reuters}} On 20 January 2025, Chinese Premier Li Qiang invited Wenfeng to his symposium with experts and asked him to provide opinions and suggestions on a draft for comments of the annual 2024 government work report.{{Cite web|last=江钰涵|date=22 January 2025|title=量化巨头幻方创始人梁文锋参加总理座谈会并发言, 他还创办了"AI界拼多多"|url=https://finance.sina.com.cn/jjxw/2025-01-22/doc-inefuxsi7314244.shtml|access-date=31 January 2025|work=Sina Corp |archive-date=26 January 2025|archive-url=https://web.archive.org/web/20250126142226/https://finance.sina.com.cn/jjxw/2025-01-22/doc-inefuxsi7314244.shtml|url-status=live}} On 20 February 2025, Wenfeng met with General Secretary of the Chinese Communist Party Xi Jinping, who encouraged party and state leaders to experiment with DeepSeek. Government officials responded to Xi's approval of the chatbot by reportedly using it to draft legal judgements, propose medical treatment plans, and analyze surveillance videos to search for missing persons.

Leading figures in the American AI sector had mixed reactions to DeepSeek's performance and success.{{cite magazine |last=Sherry|first=Ben|date=28 January 2025|title=DeepSeek, Calling It 'Impressive' but Staying Skeptical|url=https://www.inc.com/ben-sherry/ai-leaders-in-the-u-s-react-to-deepseek-calling-it-impressive-but-staying-skeptical/91140125|url-status=live|archive-url=https://web.archive.org/web/20250129211401/https://www.inc.com/ben-sherry/ai-leaders-in-the-u-s-react-to-deepseek-calling-it-impressive-but-staying-skeptical/91140125|archive-date=29 January 2025|access-date=29 January 2025|website=Inc.}} Microsoft CEO Satya Nadella and Altman—whose companies are involved in the United States government-backed "Stargate Project" to develop American AI infrastructure—both called DeepSeek "super impressive".{{Cite web|last=Okemwa |first=Kevin|date=28 January 2025|title=Microsoft CEO Satya Nadella touts DeepSeek's open-source AI as "super impressive": "We should take the developments out of China very, very seriously" |url=https://www.windowscentral.com/software-apps/microsoft-ceo-satya-nadella-touts-deepseeks-open-source-ai-as-super-impressive|url-status=live|archive-url=https://web.archive.org/web/20250128095125/https://www.windowscentral.com/software-apps/microsoft-ceo-satya-nadella-touts-deepseeks-open-source-ai-as-super-impressive|archive-date=28 January 2025 |access-date=28 January 2025|website=Windows Central}}{{cite news|last=Nazzaro|first=Miranda |date=28 January 2025|title=OpenAI's Sam Altman calls DeepSeek model 'impressive'|url=https://thehill.com/policy/technology/5110258-openai-rival-deepseek-ai-model/|url-status=live|archive-url=https://web.archive.org/web/20250128161932/https://thehill.com/policy/technology/5110258-openai-rival-deepseek-ai-model/|archive-date=28 January 2025|access-date=28 January 2025|work=The Hill}} Various companies including Amazon Web Services, Toyota, and Stripe are seeking to use the model in their program.{{Cite web |last=Kim |first=Eugene |date=27 January 2025 |title=Big AWS customers, including Stripe and Toyota, are hounding the cloud giant for access to DeepSeek AI models |url=https://www.businessinsider.com/aws-deepseek-customer-cloud-access-bedrock-stripe-toyota-cisco-workday-2025-1 |url-status=live |archive-url=https://web.archive.org/web/20250131205127/https://www.businessinsider.com/aws-deepseek-customer-cloud-access-bedrock-stripe-toyota-cisco-workday-2025-1 |archive-date=31 January 2025 |access-date=4 February 2025 |website=Business Insider}} When American President Donald Trump announced The Stargate Project, he referred to DeepSeek as a wake-up call{{Cite news|last1=Dou|first1=Eva|last2=Gregg|first2=Aaron |last3=Zakrzewski|first3=Cat|last4=Tiku|first4=Nitasha|last5=Najmabadi|first5=Shannon|date=28 January 2025|title=Trump calls China's DeepSeek AI app a 'wake-up call' after tech stocks slide |url=https://www.washingtonpost.com/business/2025/01/27/deep-seek-ai-markets-nvidia/|url-status=live|archive-url=https://web.archive.org/web/20250128030936/https://www.washingtonpost.com/business/2025/01/27/deep-seek-ai-markets-nvidia/|archive-date=28 January 2025|access-date=28 January 2025|newspaper=The Washington Post}} and a positive development.{{Cite web |last=Habeshian|first=Sareen|date=28 January 2025|title=Johnson bashes China on AI, Trump calls DeepSeek development "positive" |url=https://www.axios.com/2025/01/28/trump-mike-johnson-china-deepseek-ai|url-status=live|archive-url=https://web.archive.org/web/20250204083126/https://www.axios.com/2025/01/28/trump-mike-johnson-china-deepseek-ai|archive-date=4 February 2025|access-date=4 February 2025|website=Axios}}{{Cite web|last1=Karaian|first1=Jason|last2=Rennison|first2=Joe|date=27 January 2025|title=China's A.I. Advances Spook Big Tech Investors on Wall Street|url=https://www.nytimes.com/2025/01/27/business/us-stock-market-deepseek-ai-sp500-nvidia.html|url-status=live|archive-url=https://web.archive.org/web/20250130192538/https://www.nytimes.com/2025/01/27/business/us-stock-market-deepseek-ai-sp500-nvidia.html |archive-date=30 January 2025|access-date=4 February 2025|work=New York Times}}

Other leaders in the AI field, however—including Scale AI CEO Alexandr Wang, Anthropic cofounder and CEO Dario Amodei, and Elon Musk—have expressed skepticism of the app's performance or of the sustainability of its success.{{Cite web|last=Sharma|first=Manoj|date=6 January 2025 |title=Musk dismisses, Altman applauds: What leaders say on DeepSeek's disruption|url=https://www.fortuneindia.com/technology/musk-dismisses-sam-altman-applauds-heres-what-leaders-say-on-deepseeks-ai-disruption/120194|url-status=live|archive-url=https://web.archive.org/web/20250128140632/https://www.fortuneindia.com/technology/musk-dismisses-sam-altman-applauds-heres-what-leaders-say-on-deepseeks-ai-disruption/120194|archive-date=28 January 2025|access-date=28 January 2025|website=Fortune India}}{{Cite web|last=Dhawan|first=Sunil|date=28 January 2025 |title=Elon Musk 'questions' DeepSeek's claims, suggests massive Nvidia GPU infrastructure |url=https://www.financialexpress.com/business/investing-abroad-deepseek-sparks-debate-did-it-build-cutting-edge-ai-for-6-million-3728638/|access-date=28 January 2025|website=The Financial Express}}

DeepSeek's optimization of limited resources has highlighted potential limits of United States sanctions on China's AI development, including export restrictions on advanced AI chips to China.{{cite web|last1=Jiang |first1=Ben|last2=Perezi|first2=Bien|date=1 January 2025|title=Meet DeepSeek: the Chinese start-up that is changing how AI models are trained|url=https://www.scmp.com/tech/tech-trends/article/3293050/meet-deepseek-chinese-start-changing-how-ai-models-are-trained|work=South China Morning Post|access-date=1 January 2025|archive-date=22 January 2025|archive-url=https://web.archive.org/web/20250122160046/https://www.scmp.com/tech/tech-trends/article/3293050/meet-deepseek-chinese-start-changing-how-ai-models-are-trained|url-status=live}}{{Cite web |last=Shilov|first=Anton|date=27 December 2024|title=Chinese AI company's AI model breakthrough highlights limits of US sanctions |url=https://www.tomshardware.com/tech-industry/artificial-intelligence/chinese-ai-company-says-breakthroughs-enabled-creating-a-leading-edge-ai-model-with-11x-less-compute-deepseeks-optimizations-highlight-limits-of-us-sanctions|url-status=live |archive-url=https://web.archive.org/web/20241228014832/https://www.tomshardware.com/tech-industry/artificial-intelligence/chinese-ai-company-says-breakthroughs-enabled-creating-a-leading-edge-ai-model-with-11x-less-compute-deepseeks-optimizations-highlight-limits-of-us-sanctions |archive-date=28 December 2024|access-date=28 December 2024|website=Tom's Hardware}} The success of the company's AI models consequently "sparked market turmoil"{{Cite web |title=DeepSeek updates – Chinese AI chatbot sparks US market turmoil, wiping $500bn off Nvidia |url=https://www.bbc.co.uk/news/live/cjr85l2e4l4t|access-date=27 January 2025|website=BBC News|archive-date=27 January 2025|archive-url=https://web.archive.org/web/20250127165452/https://www.bbc.co.uk/news/live/cjr85l2e4l4t|url-status=live}} and caused shares in major global technology companies to plunge on 27 January 2025: Nvidia's stock fell by as much as 17–18%,{{cite news |last=Nazareth|first=Rita|date=26 January 2025|title=Stock Rout Gets Ugly as Nvidia Extends Loss to 17%: Markets Wrap |url=https://www.bloomberg.com/news/articles/2025-01-26/asia-eyes-cautious-open-as-tariffs-remain-in-focus-markets-wrap|access-date=27 January 2025|website=Bloomberg L.P.}} as did the stock of rival Broadcom. Other tech firms also sank, including Microsoft (down 2.5%), Google's owner Alphabet (down over 4%), and Dutch chip equipment maker ASML (down over 7%). A global sell-off of technology stocks on Nasdaq, prompted by the release of the R1 model, led to record losses of about $593 billion in the market capitalizations of AI and computer hardware companies;{{cite web|last1=Carew|first1=Sinéad |last2=Cooper|first2=Amanda|last3=Banerjee |first3=Ankur|date=27 January 2025|title=DeepSeek sparks global AI selloff, Nvidia losses about $593 billion of value |url=https://www.reuters.com/technology/chinas-deepseek-sets-off-ai-market-rout-2025-01-27/|website=Reuters|access-date=4 February 2025|archive-date=28 January 2025|archive-url=https://web.archive.org/web/20250128033924/https://www.reuters.com/technology/chinas-deepseek-sets-off-ai-market-rout-2025-01-27/|url-status=live}} and by the next day a total of $1 trillion of value was wiped from American stocks.

File:Deepseek login error.png

DeepSeek has been reported to sometimes claim that it is ChatGPT.{{Cite web|last1=O'Brien|first1=Matt|last2=Chan|first2=Kelvin |date=29 January 2025|title=Did DeepSeek copy ChatGPT to make new AI chatbot? Trump adviser thinks so |url=https://apnews.com/article/deepseek-ai-chatgpt-openai-copyright-a94168f3b8caa51623ce1b75b5ffcc51|access-date=4 February 2025|website=AP News}}{{Cite web|last=Ulanoff |first=Lance|date=30 January 2025|title=DeepSeek just insisted it's ChatGPT, and I think that's all the proof I need |url=https://www.techradar.com/computing/artificial-intelligence/deepseek-just-insisted-its-chatgpt-and-i-think-thats-all-the-proof-i-need|access-date=4 February 2025|website=TechRadar}} OpenAI said that DeepSeek may have "inappropriately" used outputs from its model as training data in a process called distillation.{{Cite web|last1=Collier|first1=Kevin|last2=Cui|first2=Jasmine|date=30 January 2025|title=OpenAI says DeepSeek may have 'inapproriately' used its data|url=https://www.nbcnews.com/tech/tech-news/openai-says-deepseek-may-inapproriately-used-data-rcna189872|access-date=5 February 2025|website=NBC News}} However, there is currently no method to prove this conclusively.{{Cite web |last1=Frermann|first1=Lea|last2=Cohney|first2=Shaanan|date=4 February 2025|title=OpenAI says DeepSeek 'inappropriately' copied ChatGPT – but it's facing copyright claims too |url=https://theconversation.com/openai-says-deepseek-inappropriately-copied-chatgpt-but-its-facing-copyright-claims-too-248863|access-date=11 February 2025|website=The Conversation}}

{{See also|Chinese censorship abroad|Censorship in China}}

DeepSeek's compliance with Chinese government censorship policies and its data collection practices have raised concerns over information control in the model, prompting regulatory scrutiny in multiple countries.

Reports indicate that it applies content moderation in accordance with the government's "public opinion guidance" regulations, limiting responses on topics such as the Tiananmen Square massacre and Taiwan's political status.{{Cite news|last1=Field |first1=Matthew|last2=Titcomb|first2=James|date=27 January 2025|title=Chinese AI has sparked a $1 trillion panic – and it doesn't care about free speech |url=https://www.telegraph.co.uk/business/2025/01/27/chinese-deepseek-ai-has-sparked-a-1-trillion-panic/|access-date=27 January 2025|work=The Daily Telegraph|issn=0307-1235}}{{cite web|last=Lu|first=Donna|date=28 January 2025|title=We tried out DeepSeek. It worked well, until we asked it about Tiananmen Square and Taiwan |url=https://www.theguardian.com/technology/2025/jan/28/we-tried-out-deepseek-it-works-well-until-we-asked-it-about-tiananmen-square-and-taiwan|access-date=30 January 2025|work=The Guardian|issn=0261-3077}}{{Cite news |last=Colville |first=Alex |date=April 25, 2025 |title=DeepSeek: A Tool Tuned for Social Governance |url=https://jamestown.org/program/deepseek-a-tool-tuned-for-social-governance/ |access-date=April 26, 2025 |work=China Brief |publisher=Jamestown Foundation}} DeepSeek models that have been uncensored also display a bias towards Chinese government viewpoints on controversial topics such as Xi Jinping's human rights record and Taiwan's political status.{{Cite web|last=Colville|first=Alex|date=10 February 2025|title=DeepSeeking Truth |url=https://chinamediaproject.org/2025/02/10/deepseeking-truth/ |access-date=12 February 2025|work=China Media Project}}{{Cite magazine|last=Yang|first=Ziyi|date=31 January 2025|title=Here's How DeepSeek Censorship Actually Works - And How to Get Around It|url=https://www.wired.com/story/deepseek-censorship/|access-date=12 February 2025|magazine=Wired}} However, users who have downloaded the models and hosted them on their own devices and servers have reported successfully removing this censorship.{{Cite news |last1=Booth|first1=Robert|last2=Milmo|first2=Dan|date=28 January 2025 |title=Chinese AI chatbot DeepSeek censors itself in realtime, users report |url=https://www.theguardian.com/technology/2025/jan/28/chinese-ai-chatbot-deepseek-censors-itself-in-realtime-users-report|access-date=13 February 2025|work=The Guardian|issn=0261-3077}}{{Cite web|last=Nield|first=David|date=29 January 2025 |title=Want to try DeepSeek without the privacy worries? Perplexity AI just launched it on its iOS and web apps |url=https://www.techradar.com/computing/artificial-intelligence/want-to-try-deepseek-without-the-privacy-worries-perplexity-ai-just-launched-it-on-its-ios-and-web-apps|access-date=13 February 2025 |website=TechRadar}}

File:DeepSeek when asked about Xi Jinping and Narendra Modi.png is an autocrat, DeepSeek apologises that this question is "beyond my current scope". The same question posed about Narendra Modi returns a "balanced analysis" of viewpoints.]] Some sources{{Who|date=May 2025}} have observed that the official application programming interface (API) version of R1, which runs from servers located in mainland China, uses censorship mechanisms for topics considered politically sensitive for the government of China. For example, the model may initially generate answers to questions about the 1989 Tiananmen Square massacre, persecution of Uyghurs, comparisons between Xi Jinping and Winnie the Pooh, and human rights in China, but a censorship mechanism deletes the uncensored response afterwards and replaces it with a message such as:{{Cite news|last1=Field|first1=Matthew |last2=Titcomb|first2=James|date=27 January 2025|title=Chinese AI has sparked a $1 trillion panic – and it doesn't care about free speech |url=https://www.telegraph.co.uk/business/2025/01/27/chinese-deepseek-ai-has-sparked-a-1-trillion-panic/|access-date=27 January 2025|work=The Daily Telegraph|issn=0307-1235|archive-date=27 January 2025|archive-url=https://web.archive.org/web/20250127142008/https://www.telegraph.co.uk/business/2025/01/27/chinese-deepseek-ai-has-sparked-a-1-trillion-panic/|url-status=live}}{{cite web|last=Lu|first=Donna|date=28 January 2025|title=We tried out DeepSeek. It worked well, until we asked it about Tiananmen Square and Taiwan|url=https://www.theguardian.com/technology/2025/jan/28/we-tried-out-deepseek-it-works-well-until-we-asked-it-about-tiananmen-square-and-taiwan|work=The Guardian|issn=0261-3077|access-date=30 January 2025}}{{Cite web|last=Steinschaden|first=Jakob|date=27 January 2025|title=DeepSeek: This is what live censorship looks like in the Chinese AI chatbot|url=https://www.trendingtopics.eu/deepseek-this-is-what-live-censorship-looks-like-in-the-chinese-ai-chatbot/|access-date=27 January 2025|website=Trending Topics|archive-date=27 January 2025|archive-url=https://web.archive.org/web/20250127144208/https://www.trendingtopics.eu/deepseek-this-is-what-live-censorship-looks-like-in-the-chinese-ai-chatbot/|url-status=live}}

"Sorry, that's beyond my current scope. Let's talk about something else."

The censorship mechanisms and restrictions added on top of the model's output can be removed in the open-source version of the R1 model.{{Cite web|last=Colville|first=Alex|date=10 February 2025|title=DeepSeeking Truth |url=https://chinamediaproject.org/2025/02/10/deepseeking-truth/ |access-date=12 February 2025|website=China Media Project}} If the "core Socialist values" defined by the Chinese Internet regulatory authorities are undermined in a response, or if a response undermining the reunification of mainland China and Taiwan is produced, the relevant response is deleted.{{Cite news|date=26 January 2025|title=The Guardian view on a global AI race: geopolitics, innovation and the rise of chaos |url=https://www.theguardian.com/commentisfree/2025/jan/26/the-guardian-view-on-a-global-ai-race-geopolitics-innovation-and-the-rise-of-chaos|access-date=27 January 2025|work=The Guardian|issn=0261-3077}} Locally hosted instances of R1 are still occasionally reported to provide answers consistent with Chinese Communist Party propaganda narratives. When tested by NBC News, for instance, DeepSeek's R1 initially described Taiwan as "an inalienable part of China's territory", and stated:{{Cite web|last1=Yang|first1=Angela |last2=Cui|first2=Jasmine|date=27 January 2025|title=Chinese AI DeepSeek jolts Silicon Valley, giving the AI race its 'Sputnik moment' |url=https://www.nbcnews.com/tech/tech-news/china-ai-assistant-deepseek-rcna189385|access-date=27 January 2025|work=NBC News}}

"We firmly oppose any form of 'Taiwan independence' separatist activities and are committed to achieving the complete reunification of the motherland through peaceful means."

In January 2025, however, Western researchers were able to trick DeepSeek into giving certain answers to some of these topics by requesting in its answer to swap certain letters for similar-looking numbers.

{{See also|Chinese information operations and information warfare}}

Many experts{{Who|date=May 2025}} fear that the government of China could use the AI system for foreign influence operations, spreading disinformation, surveillance, and the development of cyberweapons.{{cite web|last=Kimery|first=Anthony|date=26 January 2025|title=China's DeepSeek AI poses formidable cyber, data privacy threats|url=https://www.biometricupdate.com/202501/chinas-deepseek-ai-poses-formidable-cyber-data-privacy-threats |website=Biometric Update|access-date=27 January 2025}}{{Cite news|last1=Booth|first1=Robert|last2=Milmo|first2=Dan|date=28 January 2025|title=Experts urge caution over use of Chinese AI DeepSeek |url=https://www.theguardian.com/technology/2025/jan/28/experts-urge-caution-over-use-of-chinese-ai-deepseek|access-date=28 January 2025|work=The Guardian|issn=0261-3077}}{{Cite web|last=Hornby|first=Rael|date=28 January 2025 |title=DeepSeek's success has painted a huge TikTok-shaped target on its back|url=https://www.laptopmag.com/ai/deepseeks-success-has-painted-a-huge-tiktok-shaped-target-on-its-back|access-date=28 January 2025|work=LaptopMag|archive-date=28 January 2025|archive-url=https://web.archive.org/web/20250128142525/https://www.laptopmag.com/ai/deepseeks-success-has-painted-a-huge-tiktok-shaped-target-on-its-back|url-status=live}} An article in Wired commented that the DeepSeek online service sending data to its home country might set "the stage for greater scrutiny".{{Cite magazine|last1=Burgess|first1=Matt|last2=Newman |first2=Lily Hay|date=27 January 2025|title=DeepSeek's Popular AI App Is Explicitly Sending US Data to China |url=https://www.wired.com/story/deepseek-ai-china-privacy-data/|access-date=28 January 2025|magazine=Wired|issn=1059-1028|archive-date=27 January 2025|archive-url=https://web.archive.org/web/20250127232432/https://www.wired.com/story/deepseek-ai-china-privacy-data/|url-status=live}}

In February 2025, South Korea's data protection regulator, the Personal Information Protection Commission (PIPC), raised concerns over DeepSeek, confirming that it sent the nation's user data to the owner of TikTok (ByteDance) in China. Accordingly, the PIPC banned new downloads of the app until DeepSeek addressed these concerns. The company's representative in Korea has partially acknowledged its shortcomings in complying with local data protection laws.{{Cite news|last=Kim |first=Hyun-soo|date=18 February 2025|title=DeepSeek sent S. Korean user data to China's ByteDance: regulator |url=https://en.yna.co.kr/view/AEN20250218005300315|archive-url=https://web.archive.org/web/20250218135802/https://en.yna.co.kr/view/AEN20250218005300315|archive-date=18 February 2025|access-date=19 February 2025|work=Yonhap News}}{{Cite news|date=17 February 2025|title=New downloads of DeepSeek suspended in South Korea, data protection agency says |url=https://www.reuters.com/technology/south-koreas-data-protection-authority-suspends-local-service-deepseek-2025-02-17/|access-date=19 February 2025|work=Reuters}}

On 29 January 2025, cybersecurity researchers from Wiz Research found a significant data breach at DeepSeek. This occurred due to an improper configuration of a cloud storage database that was left available online without any security precautions.The database at oauth2callback.deepseek.com:9000 and dev.deepseek.com:9000 contained more than a million sensitive entries, which comprised chat messages, API keys, operational details from the backend, system logs, and internal metadata from 6 January 2025. The breach was consequential as it provided complete control over the database, which could lead to unauthorized access and increased privileges within DeepSeek's systems. After Wiz Research reported the issue, DeepSeek secured the database in less than an hour, eliminating any immediate risks. However, the exposure sparked serious concerns about DeepSeek's data security practices and the possibility of facing regulatory action under laws like GDPR or CCPA, particularly if the data included personal information from EU or US residents.{{Cite web |last=Daniel |first=Lars |title=DeepSeek Data Leak Exposes 1,000,000 Sensitive Records |url=https://www.forbes.com/sites/larsdaniel/2025/02/01/deepseek-data-leak-exposes--1000000-sensitive-records/ |access-date=2025-04-06 |website=Forbes}}{{Cite web |date=2025-01-29 |title=Wiz Research Uncovers Exposed DeepSeek Database Leaking Sensitive Information, Including Chat History {{!}} Wiz Blog |url=https://www.wiz.io/blog/wiz-research-uncovers-exposed-deepseek-database-leak |access-date=2025-04-06 |website=wiz.io}}

In April 2025, South Korea's Personal Information Protection Commission announced that DeepSeek had transferred personal information on over a million South Koreans to China without consent.{{Cite web |last=Kang |first=Taejun |date=2025-04-24 |title=China's DeepSeek transferred South Korean user info overseas: Seoul regulator |url=https://www.rfa.org/english/china/2025/04/24/china-deep-seek-south-korea-user-data/ |access-date=2025-04-24 |website=Radio Free Asia}} In June 2025, the Italian Competition Authority opened an investigation into DeepSeek over false information risks.{{Cite news |date=June 16, 2025 |title=Italy regulator probes DeepSeek over false information risks |url=https://www.reuters.com/world/china/italy-regulator-opens-probe-into-chinas-deepseek-2025-06-16/ |access-date=June 16, 2025 |work=Reuters}}

The Wall Street Journal reported that the DeepSeek app produces instructions for self-harm and dangerous activities more often than its American competitors.{{Cite web|last=Schechner |first=Sam|date=8 February 2025|title=DeepSeek Offers Bioweapon, Self-Harm Information|url=https://www.wsj.com/tech/ai/china-deepseek-ai-dangerous-information-e8eb31a8|url-access=subscription|url-status=live|archive-url=https://archive.today/20250209235331/https://www.wsj.com/tech/ai/china-deepseek-ai-dangerous-information-e8eb31a8|archive-date=9 February 2025|access-date=12 February 2025|website=The Wall Street Journal}} Security researchers{{Who|date=May 2025}} have found that DeepSeek sends data to a cloud platform affiliated with ByteDance.{{Cite web|last=Goodin|first=Dan|date=6 February 2025 |title=DeepSeek iOS app sends data unencrypted to ByteDance-controlled servers|url=https://arstechnica.com/security/2025/02/deepseek-ios-app-sends-data-unencrypted-to-bytedance-controlled-servers/|access-date=9 February 2025|website=Ars Technica}}{{Cite web|date=6 February 2025|title=Experts Flag Security, Privacy Risks in DeepSeek AI App |url=https://krebsonsecurity.com/2025/02/experts-flag-security-privacy-risks-in-deepseek-ai-app/|access-date=9 February 2025|work=Krebs on Security}}

In February 2025, sources claimed that DeepSeek began considering raising external funding for the first time, with Alibaba and Chinese state funds expressing interest in investing in DeepSeek.{{Cite web|date=February 20, 2025|title=DeepSeek's outside funding draws interest from Alibaba, Chinese state funds, The Information reports |url=https://www.reuters.com/technology/artificial-intelligence/alibaba-chinese-state-funds-interested-investing-deepseek-information-reports-2025-02-19/|access-date=February 25, 2025 |website=Reuters}}{{Cite web|date=February 19, 2025 |title=DeepSeek Weighs Raising Outside Money For First Time |url=https://www.tradingkey.com/news/more-news/250387930-reuters |website=TradingKey|access-date=February 25, 2025}} On 21 February, DeepSeek announced plans to release key codes and data to the public starting the following week.{{Cite web |date=February 21, 2025|title=Day 0: Warming up for #OpenSourceWeek! |url=https://x.com/deepseek_ai/status/1892786555494019098|publisher=X|access-date=February 25, 2025}}{{Cite web |last=Rai|first=Saritha|date=February 21, 2025|title=DeepSeek Promises to Share Even More AI Code in a Rare Step |url=https://www.bloomberg.com/news/articles/2025-02-21/deepseek-promises-to-share-even-more-ai-code-in-a-rare-step|website=Bloomberg LP|access-date=February 25, 2025}}

Many countries have raised concerns about data security and DeepSeek's use of personal data. On 28 January 2025, the Italian data protection authority announced that it is seeking additional information on DeepSeek's collection and use of personal data.{{Cite news|date=28 January 2025|title=Italy regulator seeks information from DeepSeek on data protection |url=https://www.reuters.com/technology/artificial-intelligence/italy-regulator-seeks-info-deepseek-data-protection-2025-01-28/|access-date=28 January 2025|website=Reuters}} On the same day, the United States National Security Council announced that it had started a national security review of DeepSeek,{{Cite news |last1=Shalal|first1=Andrea|last2=Shepardson|first2=David|date=28 January 2025|title=White House evaluates effect of China AI app DeepSeek on national security, official says |url=https://www.reuters.com/technology/artificial-intelligence/white-house-evaluates-china-ai-app-deepseeks-affect-national-security-official-2025-01-28/|access-date=28 January 2025|website=Reuters}} and the United States Navy instructed all its members not to use DeepSeek due to "security and ethical concerns."{{Cite web |last=Field|first=Hayden|date=28 January 2025|title=U.S. Navy bans use of DeepSeek due to 'security and ethical concerns' |url=https://www.cnbc.com/2025/01/28/us-navy-restricts-use-of-deepseek-ai-imperative-to-avoid-using.html|access-date=6 February 2025 |website=CNBC}} On 30 January 2025, Italy's data protection authority, ordered DeepSeek to block its chatbot in the country after the Chinese artificial intelligence startup failed to address the regulator's concerns over its privacy policy. Italian users who downloaded DeepSeek on mobile devices before the ban reported being able to use the chatbot and having access to the web version of DeepSeek after the ban.{{Cite news|date=31 January 2025|title=Italy's regulator blocks Chinese AI app DeepSeek on data protection |url=https://www.reuters.com/technology/artificial-intelligence/italys-privacy-watchdog-blocks-chinese-ai-app-deepseek-2025-01-30/|access-date=31 January 2025|website=Reuters}}

Three days later, South Korea's Personal Information Protection Commission opened an inquiry into DeepSeek's use of personal information;{{Cite news |date=31 January 2025|title=Taiwan says government departments should not use DeepSeek, citing security concerns |url=https://www.reuters.com/technology/artificial-intelligence/taiwan-says-government-departments-should-not-use-deepseek-citing-security-2025-01-31/|access-date=31 January 2025 |website=Reuters}} the Dutch Data Protection Authority launched an investigation of DeepSeek;{{Cite news|date=31 January 2025|title=Dutch privacy watchdog to launch investigation into China's DeepSeek AI|url=https://www.reuters.com/technology/artificial-intelligence/dutch-privacy-watchdog-launch-investigation-into-chinas-deepseek-ai-2025-01-31/|access-date=1 February 2025|website=Reuters}} Taiwan's digital ministry advised its government departments against using the DeepSeek service to "prevent information security risks"; and Texas Governor Greg Abbott issued a state ban on government-issued devices for DeepSeek, along with Xiaohongshu and Lemon8.{{Cite web|last=Lathan|first=Nadia|date=31 January 2025|title=Texas governor orders ban on DeepSeek, RedNote for government devices|url=https://apnews.com/article/texas-deepseek-apps-ban-3828a4743e9919398dfac0ba9d4a5c25|access-date=1 February 2025|website=Associated Press}}

In February 2025, access to DeepSeek was banned on the New South Wales Department of Customer Service's devices.{{Cite web |last=Williams|first=Tom|date=4 February 2025|title=NSW Govt blocks access to DeepSeek AI |url=https://ia.acs.org.au/article/2025/nsw-govt-blocks-access-to-deepseek-ai.html|access-date=4 February 2025|website=Information Age|archive-date=4 February 2025|archive-url=https://web.archive.org/web/20250204130742/https://ia.acs.org.au/article/2025/nsw-govt-blocks-access-to-deepseek-ai.html|url-status=live}} That same month, Australia, South Korea, and Canada banned DeepSeek from government devices,{{Cite web |last=Gerken|first=Tom|date=4 February 2025|title=Australia bans DeepSeek on government devices over security risk |url=https://www.bbc.com/news/articles/c8d95v0nr1yo|access-date=4 February 2025|website=BBC}}{{Cite web |last=Lim|first=Lionel|date=6 February 2025|title=South Korea's government is the latest to block China's DeepSeek on official devices, following Australia and Taiwan |url=https://fortune.com/asia/2025/02/06/south-korea-blocks-deepseek-government-devices-china-ai-taiwan-australia/|access-date=6 February 2025|website=Fortune}}{{Cite news|last=Christopher|first=Nardi|date=6 February 2025 |title=Federal government bans Chinese AI startup DeepSeek on public service devices|url=https://nationalpost.com/news/canada/federal-government-bans-chinese-ai-startup-deepseek-on-public-service-devices |access-date=7 February 2025|work=National Post}} and South Korea suspended new downloads of DeepSeek due to risks of personal information misuse.{{Cite web|first=Sang-Seo|last=Lee|date=17 February 2025|title=Personal Information Protection Commission suspends new services of Deepseek due to insufficient personal information policy |url=https://m.yna.co.kr/view/AKR20250217066600530|access-date=17 February 2025|website=Yonhap|language=Korean}} In March 2025, the U.S. Department of Commerce prohibited DeepSeek on their government devices, due to the danger of disclosing proprietary information with the Chinese Communist Party.{{Cite news |title=US Commerce department bureaus ban China's DeepSeek on government devices, sources say |url=https://www.reuters.com/technology/artificial-intelligence/us-commerce-department-bureaus-ban-chinas-deepseek-government-devices-sources-2025-03-17/ |archive-url=http://web.archive.org/web/20250323140252/https://www.reuters.com/technology/artificial-intelligence/us-commerce-department-bureaus-ban-chinas-deepseek-government-devices-sources-2025-03-17/ |archive-date=2025-03-23 |access-date=2025-04-10 |work=Reuters}}

• Zhejiang University

{{Portal|Free and open-source software|China}}

{{notelist}}

{{reflist|refs=

}}

• {{Official website}}

{{AI-based chatbots}}

{{Generative AI}}

{{Artificial intelligence navbox}}

Category:2025 software

Category:Chatbots

Category:Generative pre-trained transformers

Category:Interactive narrative

Category:2025 in artificial intelligence