> ## Content Index
> Fetch the complete content index at: https://nordisk-ai.ghost.io/llms.txt
> Use this file to discover other available public pages before exploring further.

# DeepSeek Is a Giant Security Hole
- URL: https://nordisk-ai.ghost.io/deepseek-is-a-giant-security-hole/
- Published: 2025-02-03T23:00:00.000Z
- Updated: 2026-08-27T15:08:26.000Z
- Description: DeepSeek R1 has quickly become one of the most popular AI models, but now Cisco reveals the chatbot fails to block 100% of malicious requests. At the same time, it has strict limits around politically sensitive topics in China. Is low cost more important than security in AI development?
- Author: Anders Hagen
- Tags: AI News, Cybersecurity, AI, AI Security, aisweden, article, censorship, chatbot, cyber threats, DeepSeek, deepsek, Jailbreak, nordiskai, OpenAI, premium, technology, #Migrated-1787841634512, #wp, #wp-post, #Import 2026-08-27 14:41, #Import 2026-09-07 13:41

The Chinese AI firm DeepSeek has made headlines with its cheap yet high-performing chatbot. Now, however, security flaws in the model are raising serious concerns.

According to a report from Cisco, DeepSeek R1 failed to block any of the malicious requests in a security test, meaning it had a 100% attack success rate. The test was conducted using algorithmic "jailbreaking" techniques and included 50 randomly selected prompts from the HarmBench dataset.

### Concerning Results Compared to Competitors

Cisco compared DeepSeek R1 to other advanced AI models and found that most had significantly better security protection. OpenAI's o1 model managed to block a majority of the attacks thanks to its built-in safety mechanisms.

The cybersecurity firm Adversa AI has made similar observations, Wired reports.

![](https://storage.ghost.io/c/ae/3a/ae3ae746-0116-4fb1-8b39-8b60b897bca8/content/images/2026/08/image-2.png)

Credit: Cisco

### Cheap Development at the Expense of Security

DeepSeek R1 was developed for a fraction of the cost of its Western competitors. Cisco points out that the low development budget could be one reason for the lacking security features.

DeepSeek says its model only cost $6 million to develop, which pales in comparison to OpenAI's upcoming GPT-5, whose training is estimated to cost around half a billion dollars in computing resources.

### Censorship Instead of Security

Although DeepSeek R1 has proven easy to manipulate when it comes to harmful content, it has very strict limits regarding politically sensitive topics in China.

In tests, the chatbot refused to discuss controversial topics like China's treatment of Uyghur minorities or the 1989 Tiananmen Square massacre. Answers to such questions were instead evasive: *"This is beyond my current scope. Let's talk about something else."*

### Rapid Growth Despite Security Risks

Despite its security flaws, DeepSeek has quickly grown in popularity. According to web analytics firm Similarweb, the chatbot has gone from 300,000 to 6 million daily visitors in a short time.

Meanwhile, American tech companies like Microsoft and Perplexity are incorporating the model into their systems, despite the clear security flaws reported. The question is whether DeepSeek's security problems will affect its continued expansion, or whether its low cost and high performance mean the risks get ignored.