Chinese State Hackers Are Scaling Attacks With DeepSeek
Chinese state-sponsored hacking groups have doubled their attack volume by handing off reconnaissance work and malware development to DeepSeek, according to new research from Taiwanese threat intelligence firm TeamT5. The finding is notable not because a sophisticated new tool has appeared, but because the opposite is true: hackers are getting more done with a cheap, weakly restricted AI model rather than a cutting-edge one.
TeamT5's research describes a pattern where state-linked groups delegate the repetitive, time-consuming parts of an attack, mapping out a target's network, drafting exploit code, and writing malware variants, to DeepSeek rather than handling that work manually. Because DeepSeek imposes minimal safety restrictions and is inexpensive to run at scale, hacking teams can generate large volumes of reconnaissance and malicious code without the friction that more heavily guarded commercial AI models typically introduce. The result, per TeamT5, is that these groups are now conducting more than twice as many attacks as before they began using the tool this way.
Why Weak Guardrails Matter More Than Raw Power
The conventional assumption in cybersecurity circles has been that more capable AI models pose the bigger threat, since they can reason through complex attack chains and produce more convincing outputs. TeamT5's findings flip that expectation. What matters most to attackers running large-scale operations isn't necessarily the smartest model, it's the one that will do what's asked without pushback and won't bankrupt the operation at volume. DeepSeek fits both criteria: it's relatively capable, it's cheap to query repeatedly, and it doesn't apply the same level of content filtering that Western AI labs have built into their commercial products.
This isn't the first time researchers have documented state-linked Chinese actors leaning on AI tools to accelerate operations. Earlier reporting on Chinese hackers using open-source AI in an attack on Taiwan described what researchers called a fully autonomous, end-to-end AI-driven cyberattack aimed at Taiwan's government. TeamT5's new report suggests that pattern isn't an isolated incident but part of a broader shift toward AI-assisted operations becoming standard practice for these groups, with DeepSeek increasingly serving as the workhorse behind the scenes.
The Privacy Stakes for Everyday Users and Organizations
Doubling attack volume doesn't necessarily mean doubling sophistication, but it does mean more targets get hit, more reconnaissance gets done on more networks, and more custom malware variants get produced to slip past defenses. For organizations, that translates into a higher baseline probability of being scanned, probed, or targeted, even if they aren't a headline-grabbing target. For individuals, the concern is less direct but still real: state-linked groups that use AI to scale reconnaissance often aren't just going after government networks. Journalists, activists, and researchers have historically been targets of Chinese state-linked digital espionage, as documented in reporting on a China-backed espionage campaign targeting journalists and activists. Tools that make reconnaissance and malware creation cheaper and faster lower the barrier for expanding that kind of targeting to a wider pool of people.
The use of a consumer-facing AI model like DeepSeek also raises a separate privacy question: how much data flows through these systems, and who might be able to access query logs or usage patterns. While TeamT5's research focuses on how hackers are using the tool offensively, the broader implication is that AI platforms with lax safety controls are becoming embedded in threat actor workflows, and that's a trend security teams will need to track closely going forward.
What This Means For You
Most readers won't be direct targets of Chinese state-sponsored hacking groups, but the broader trend still matters. AI-assisted attacks tend to increase the sheer volume of phishing attempts, scanning activity, and malware variants circulating online, which raises the odds that basic security hygiene gets tested more often. Journalists, activists, researchers, and anyone working on topics related to China should be especially mindful, given the documented history of targeted espionage campaigns against these groups.
Practical steps remain the same regardless of how attacks are generated: keep software and operating systems patched, use strong and unique passwords with a password manager, enable multi-factor authentication wherever possible, and be skeptical of unsolicited messages or attachments, even ones that look unusually polished. AI tools can make malware and phishing content more convincing and more abundant, but they don't change the fundamentals of what keeps accounts and devices secure.
As AI models with weak safety controls continue to lower the cost of running large-scale cyber operations, expect more threat intelligence firms to report similar findings. Staying informed about how these tools are being used, and maintaining consistent security practices, remains the most reliable defense available to individuals and organizations alike.
