DeepSeek limits new accounts amid cyberattack

Chinese AI startup grapples with consequences of sudden popularity

Updated China's DeepSeek, which shook up American AI makers with the debut of its V3 and reasoning-capable R1 LLM families, has limited new signups to its web-based interface to its models due to what's said to be an ongoing cyberattack.

"Due to large-scale malicious attacks on DeepSeek's services, we are temporarily limiting registrations to ensure continued service," the biz said in a note on its status page. "Existing users can log in as usual. Thanks for your understanding and support."

The incident appears to have begun around 2133 China Standard Time (CST) on Monday, January 27, or around 0533 PST here on the US West Coast ‒ and was ongoing at the time this article was filed. Sign-ups and logins via Google's single-sign-on appeared to be working at 1030 PST.

DeekSeek did not immediately respond to a request for comment.

The outfit's AI app for iOS, DeepSeek – AI Assistant, is presently the top free download in Apple's US App store, just above OpenAI's ChatGPT.

The Chinese AI firm released DeepSeek R1 as an open source LLM last week, following its parent V3 model, claiming reasoning capabilities that rival OpenAI's GPT-o1 in a number of benchmarks.

Both can be used locally or via the web or DeepSeek's apps for free, or cloud API, though as stated above, signups for the web chatbot interface at least have been limited. Bear in mind, if you use R1 or V3 using DeepSeek's online services, such as the web or app, your conversations and data are stored in China.

Having allegedly built V3 using just $5.58 million [PDF], significantly less than Western AI firms, investors in companies like Nvidia have begun to wonder whether they need to revise their financial assumptions. The result has been a selloff in AI stocks.

The viability of open source models has long been a concern among commercial AI firms with proprietary models. AI skeptics such as Gary Marcus have previously questioned the valuation of firms like OpenAI when corporations like Meta have been giving away open source models, namely Llama, at no charge.

We will update this story as it develops. ®

Updated to add at 1030 PST, 1830 UTC

As noted above, you may be able to login at least with a Google account by now, or you may be able to get through with a regular account. Also, DeepSeek has released a family of openly available models called Janus Pro that it claims can beat OpenAI's DALL-E 3 image generator in benchmarks.