Chinese AI Giant DeepSeek Hit by Cyber Attack Impacting Registrations

• DeepSeek announced that the company started investigating a large-scale cyber attack on Tuesday.
• The company has put a halt on new registrations while the security issue remains unresolved.
• Several status updates posted in the past two days reveal problems arising from DeepSeek’s API.

China-based AI company DeepSeek has been hit by a cyber attack. The company confirmed the intrusion on its daily status page, with the latest update mentioning a “large-scale malicious attack” targeting its services. 

The DeepSeek security incident has temporarily halted registrations. Existing users are not affected and can log in as usual. The source of the attack and damage remains unknown at this moment.

Based on the status updates page, DeepSeek web services were unavailable on January 14. However, the issue was resolved the same day. On January 26, DeepSeek R1 API experienced issues, which were also marked resolved later that day.

A series of incidents were reported on January 27 related to the following:

1. Logging in or Signing up on DeepSeek - Later resolved.
2. DeepSeek API service unavailable - Later resolved.
3. DeepSeek V3 API services issues - Later resolved. 
4. DeepSeek R1 API services issues - Later resolved.

Later on the same day, the company updated the page, citing degraded performance with reference to DeepSeek API, which remains under review. 

TechNadu reported about the latest AI model series, the DeepSeek R1, being cloned by cybercriminals – with eight phishing websites impersonating the tool. 

The cybercrime threat intelligence firm KELA flagged several issues with the DeepSeek R1 model that produced malicious outputs, including ransomware deployment. 

Although AI models are being trained not to share malicious content, researchers managed to jailbreak and retrieve sensitive content like detailed instructions for creating toxins and explosive devices. 

They also found that the model could be made to share fabricated information like emails, phone numbers, salaries, and nicknames of senior OpenAI employees.