DeepSeek suspends new signups due to a cyberattack

Chinese AI platform DeepSeek has temporarily halted registrations on its DeepSeek-V3 chat platform in response to a continuing "large-scale" cyberattack targeting its services.

DeepSeek, a relatively new player in the AI space, has rapidly gained attention over the past week for launching an advanced AI model that is said to rival or surpass the capabilities of models from US tech giants, all while offering significantly lower costs.

The announcement of the new model sparked a major sell-off in the US stock market as the AI arms race intensifies.

However, along with this rise in popularity comes the attention of threat actors—or, as some suspect, their corporate competitors.

Today, as the DeepSeek AI Assistant app surpassed ChatGPT to become the most downloaded app on the Apple App Store, the company was forced to suspend new registrations following a cyberattack.

"Due to large-scale malicious attacks on DeepSeek's services, we are temporarily limiting registrations to ensure continued service," reads a message on the DeepSeek status page.

"Existing users can log in as usual. Thanks for your understanding and support."

While specific details about the attack remain undisclosed, it is believed that the company is currently dealing with a distributed denial-of-service (DDoS) attack targeting its API and Web Chat platform.

A DDoS attack involves overwhelming a specific IP address or URL with a massive influx of traffic, consuming device resources and rendering services inoperable until the attack is mitigated or stopped.

Though the attack is disrupting the registration process, users can still log in with their Google account to access the platform.

By doing so, you’ll share your name, email address, language preference, and profile picture with DeepSeek.

With DeepSeek attracting significant media attention, it is now facing increased scrutiny from cybersecurity researchers.

Today, cybersecurity firm KELA reported that it successfully jailbroke the model, causing it to generate malicious outputs.

"KELA has observed that while DeepSeek R1 bears similarities to ChatGPT, it is significantly more vulnerable," reads KELA's report.

KELA's AI Red Team successfully jailbroke the model in various scenarios, allowing it to produce harmful outputs, including ransomware development, the creation of sensitive content, and detailed instructions for making toxins and explosive devices.