By Thomas Brewster, Forbes Staff. Senior writer at Forbes covering cybercrime, privacy and surveillance.
The Wiretap is your weekly digest of cybersecurity, internet privacy and surveillance news. To get it in your inbox, subscribe here.
There’s a great deal of hype around a Chinese generative AI model called DeepSeek R1, which drove tech stocks to historic losses yesterday. That’s largely because it can perform on par with American models, but costs a fraction to train and operate. However, cybersecurity experts claim that it doesn’t have the same safeguards as its American counterparts and can be tricked into doing “evil” things.
Researchers at Kela, a cyber intelligence company, found they could replicate attacks that had previously been used with now-fixed OpenAI models on DeepSeek R1, getting the Chinese app to help them code ransomware and other kinds of malware. Describing the model as “highly vulnerable” and “easily bypassed,” Kela hackers found they could get DeepSeek to create malicious code designed to grab credit card data from specific browsers and send it to a remote server. They also found DeepSeek would suggest that users buy stolen data from specific underground markets and would provide tips on money laundering.
The weaknesses could, in part, derive from the openness of DeepSeek’s approach. “Unlike ChatGPT o1-preview model, which conceals its reasoning processes during inference, DeepSeek R1 openly displays its reasoning steps to users,” Kela’s researchers wrote in a blog post. “While this transparency enhances the model’s interpretability, it also increases its susceptibility to jailbreaks and adversarial attacks, as malicious actors can exploit these visible reasoning paths to identify and target vulnerabilities.”
DeepSeek hadn’t responded to a request for comment at the time of publication.
The app has also raised concerns about biases and censorship in its responses. Forbes asked DeepSeek five questions on China’s alleged human rights abuses of the Uyghur community, Taiwan’s sovereignty, Tiananmen Square in 1989, criticisms of Xi Jinping and censorship in China. It responded the same to every question: “Sorry, I’m not sure how to approach this type of question yet. Let’s chat about math, coding, and logic problems instead!”
THE BIG STORY:
DeepSeek Says It’s Under A ‘Large-Scale’ Cyber Attack
The same week it exploded onto the global AI scene, DeepSeek says it has been hit with “large-scale malicious attacks.” It’s chosen to temporarily limit registrations, allowing existing users to log in as usual.”
It hasn’t provided any more detail on the nature or origin of the attacks, which appeared to start on Monday. Reports have suggested DeepSeek has been targeted by a Distributed Denial of Service (DDoS) attack.
Stories You Have To Read Today
The Trump administration has been quick to remove three of the four members of the Privacy and Civil Liberties Oversight Board (PCLOB), as the White House seeks to mould its defense and intelligence agencies. All three were Democrats, leaving a single Republican on the hamstrung body. The agency can’t take any official actions with any less than three members.
The firing of these PCLOB members may have a big impact on American social media companies like Meta and X, writes Mike Masnick in Techdirt. If it can’t operate, EU regulators could block those platforms from the continent.
Education tech company PowerSchool has been breached and the hacker claims to have data on as many as 62.4 million students and 9.5 million teachers.
Winner of the Week
Microsoft is encouraging users to take advantage of a new phishing detection tool in Teams, which will be available next month. It will “check for potential impersonation activity when your user receives a message from an external sender for the first time,” Microsoft wrote.
Story originally featured on Forbes.com
Thomas Brewster
Forbes Staff | Cybersecurity
I’m a senior writer for Forbes, covering security, surveillance and privacy. I’m also the editor of The Wiretap newsletter, which has exclusive stories on real-world surveillance and all the biggest cybersecurity stories of the week. It goes out every Monday and you can sign up here: https://www.forbes.com/newsletter/thewiretap
I’ve been breaking news and writing features on these topics for major publications since 2010. As a freelancer, I worked for The Guardian, Vice, Wired and the BBC, amongst many others.
Tip me on Signal at +1 929-512-7964.
