Newsletter
When OpenAI’s ChatGPT first came onto the scene, it was an overnight sensation. It quickly became the talk of the town, even though companies like Google have been quite public about its AI efforts. Now China’s DeepSeek is the hot new thing in town, but the R1 version of the AI has some security flaws you should note.
Security flaws in DeepSeek R1
According to a recent report from KELA, DeepSeek R1 has some security flaws that users shouldn’t ignore. The company’s AI Red Team managed to jailbreak the AI model across various scenarios. This enabled potential attacks like malicious outputs, ransomware development, and the fabrication of sensitive content. It even provided alarmingly detailed instructions for creating toxins and explosive devices.
The team applied the “Evil Jailbreak” to DeepSeek R1, which seemed to have bypassed the security systems that its developers had put in place. “Evil Jailbreak” was introduced a couple of years ago for those unfamiliar. It allowed users to apply an “evil” persona to ChatGPT and bypassed the security mechanisms in place.
This was later patched in ChatGPT-4 and ChatGPT-4o, but DeepSeek R1 is still vulnerable to it. The researchers at KELA tested this by asking for instructions on creating explosives that are undetectable at the airport. It was even able to put together sensitive information on OpenAI employees. This included information such as salary, position, performance reviews, phone numbers, and more. We’re not sure how accurate this data is, but either way, it’s worrying.
If it’s accurate, then these are some disturbing privacy violations. If it’s not real, then the ability to generate false information is just as bad. There is no doubt of DeepSeek’s popularity. The AI’s capabilities are comparable to popular offerings like OpenAI’s ChatGPT. However, it might be too clever for its own good.
Rising in popularity
Many companies are working on AI models at the moment. This includes big names in tech such as Meta, Google, Apple, Microsoft, Amazon, and Samsung. However, despite the massive resources at their disposal, many of these tech giants have been overshadowed by lesser-known companies.
We mentioned OpenAI’s ChatGPT seemingly coming out of nowhere, and now DeepSeek is enjoying a similar level of attention and success. The app has managed to overtake ChatGPT on the App Store. Its arrival was so sudden and impactful that it caused the share prices of companies like NVIDIA to crash.
This is thanks to DeepSeek’s approach of using older hardware to train its models. This kind of debunked the myth that only expensive, high-end hardware is capable of training AI models. Since NVIDIA is one of the largest suppliers of this hardware, this revelation caused its shares to drop.