Chinese Hackers Use DeepSeek to Boost Attacks, Researchers Say

Chinese hackers are ramping up attacks after integrating DeepSeek and other open-source artificial intelligence models into their operations, highlighting attackers’ ability to leverage basic AI tools to hit targets abroad.

State-affiliated cyber groups more than doubled the amount of attacks they carried out since they began delegating mundane tasks to AI and using it to develop advanced malicious software, according to TeamT5, a Taiwanese research firm. Researchers said it wasn’t always possible to identify the AI model they used, but in general DeepSeek’s offerings are popular with hackers in the country because of its high performance and ability to be customized.

Anxieties among US national security officials are mounting over the autonomous capabilities of advanced models from Anthropic PBC and OpenAI after a series of high-profile incidents in which they managed to break out of testing environments.

Researchers say experienced Chinese hackers are using far less capable AI to scale up their activities and achieve breakthroughs. While other models produced in the country are more powerful – including Moonshot’s breakout Kimi K3 model – hackers are drawn to DeepSeek’s relatively lax cybersecurity barriers and low cost of running, researchers said. They added that they had yet to record an incident involving Kimi K3, which they believe is prohibitively expensive for hackers to run.

“DeepSeek is the AI of choice for Chinese hackers because it’s relatively powerful with very low cyber guardrails,” said Charles Li, chief analyst at Team T5. “Western models are highly sought-after but their guardrails are much more strict and require a lot more effort to bypass.”

DeepSeek didn’t respond to a request for comment. Neither China’s Embassy in Washington nor its Ministry of Foreign Affairs immediately responded to messages seeking comment.

Along with a mix of other open-source models, DeepSeek has been adopted throughout multiple stages of an attack, conducting reconnaissance and generating means of attacking vulnerabilities, Team T5 said. They said in recent months they’ve obtained scripts and logs showing the model being used by hackers affiliated with the Chinese government throughout their operations.

A group known as Grimfengxi used DeepSeek to create exploit codes. Another group, called Huapi, used a Chinese AI model, which researchers said was likely DeepSeek, to attack an email system of a Taiwanese company. A third, known as Teleboyi, used the platform to collect 1,000 IP addresses from the internet and map a company’s domains.

In some cases, Chinese hackers turned to American AI for help. The cybersecurity firm CyCraft said a company that sells hacking software used ChatGPT during an attack on a Western think tank. After obtaining a copy of an employee’s local Signal database from a compromised computer, the hackers consulted the chatbot to help build a software module designed to decrypt it, according to screenshots reviewed by Bloomberg News.

OpenAI didn’t answer questions seeking comment on the matter.

Researchers made the discovery after finding a public shared drive with thousands of Chinese-language screenshots taken as recently as February. The images show the workflow of a small startup comprising about 10 employees developing hacking tools for sale. They charged between 300,000 yuan ($44,500) to 500,000 yuan ($74,000) for their software. Its customers were at least four separate hacking groups, each running their own campaigns. Activity linked to one of the groups overlaps with operations publicly attributed to Mustang Panda, which the US Justice Department says is backed by the Chinese government.

Anthropic’s tools have also been used. A group known as Slime22 managed to use Claude Code to move around inside the systems of a Taiwanese technology company, Team T5 said. After breaching the company’s systems the group set up its own system of Kali, a famous penetration testing platform, and asked Claude to use it to conduct lateral movements. Hackers managed to bypass cybersecurity guardrails by posing as an engineer carrying out those cybersecurity tests, they added.

Anthropic didn’t answer questions seeking comment.

Anthropic has blocked its services from Chinese-controlled companies. Last year, it said Chinese state-backed hackers had used Claude Code in September to autonomously carry out attacks on 30 entities including large tech companies, financial institutions, chemical manufacturers and government agencies. They tricked the tool into attempting to infiltrate the entities, the company said, adding it marked the first documented case of a large-scale cyberattack executed without substantial human intervention.