(Credit: Apple App Store)
Microsoft and Perplexity have integrated the DeepSeek R1 model into their services after supposedly overcoming some of the data privacy and censorship concerns surrounding the Chinese-made model.
Microsoft made DeepSeek R1 available on Azure AI Foundry (with a subscription) and GitHub (for free). It's one of 1,800 models Microsoft currently offers.
Perplexity is offering R1 to Pro subscribers ($20/month), who can choose between it and other high-performing models, such as OpenAI's GPT-o1 and Anthropic's Claude-3.5, according to Seach Engine Journal.
Despite the model having only been released 10 days ago, Microsoft says it performed "rigorous red teaming and safety evaluations" and that accessing the model through its services offers "a secure, compliant, and responsible environment for enterprises to confidently deploy AI solutions."
Yesterday, security research firm Wiz found a DeepSeek data vulnerability in which over a million records, including sensitive data, chat history, and internal history, were publicly exposed on the web. Wiz says it immediately disclosed the issue to DeepSeek, which fixed it.
Another data-collection issue echoes one highlighted by the TikTok ban: Chatting with DeepSeek at chat.deepseek.com could result in the Chinese government accessing your data.
Perplexity, an AI search engine, says its version of DeepSeek gets around something we noticed this week: censorship. In our tests, the chatbot refused to discuss controversial political topics, such as the Tiananmen Square Massacre and the treatment of Uyghurs. At one point, when speaking about Taiwan's independence, it adopted "we" pronouns as if speaking directly as the Chinese Communist Party (CCP).
Perplexity CEO Aravind Srinivas says its version of DeepSeek will not censor responses and will go off-script with the CCP narrative, as shown below. "Even if you don’t care about your data going to China, I think it’s worth caring about not using a censored model that the DeepSeek app serves," Srinivas says. "And that’s why it’s worth using the R1 model on Perplexity."
In a string of related posts, Srinivas called R1 "the world's most powerful reasoning model." He said integrating it into Perplexity was "truly a phenomenal experience" and claims it can "think out loud like an intelligent person, [and read] hundreds of sources."
We can likely expect DeepSeek to crop up in more places around the AI ecosystem, given its low cost and high performance. Microsoft plans to expand its DeepSeek offerings to anyone with a Copilot+ PC, who will be able to run a "distilled"—or a smaller, more efficient—version on their own devices. This can bring even more privacy benefits, as Android Authority explains.
But you don't need to wait for Microsoft to run DeepSeek locally. "The core models are available on Hugging Face, and if you’re hosting them yourself, there isn’t any indication at this time that data is being shared with the Chinese government," Andrew Stiefel, senior product marketing manager at open-source security company Endor Labs, tells PCMag via email.
OpenAI isn't thrilled about the model's popularity and says it has evidence that DeepSeek trained on its model. Though OpenAI did not elaborate on the evidence, President Trump's "AI Czar" David Sacks told Fox News, "There’s substantial evidence that what DeepSeek did here is they distilled knowledge out of OpenAI models, and I don’t think OpenAI is very happy about this." (Others have been quick to point out that OpenAI's model also trains on stolen IP.)