Chinese state-backed hacking groups have more than doubled their attack output since folding AI tools into their workflows, according to Taiwanese security firm TeamT5. The finding, reported via Bloomberg, points to DeepSeek as the model of choice — not because it’s the most capable, but because, as TeamT5 chief analyst Charles Li put it, “it’s relatively powerful with very low cyber guardrails.”
That distinction matters. Western frontier models have baked-in safety layers that flag or refuse malicious prompts. DeepSeek’s lighter restrictions mean Chinese operators can use it for the unglamorous grunt work of espionage — writing exploit code, mapping domains, collecting IP addresses — without tripping alarms. TeamT5 documented specific cases: the group Grimfengxi used DeepSeek to write exploit code, Huapi relied on a Chinese model likely to be DeepSeek, and Teleboyi used the platform for reconnaissance tasks. ChatGPT appeared in at least one case, with security firm CyCraft finding evidence it helped build a decryption module for a Signal database.
The shift isn’t limited to Chinese models. A group called Slime22 used Anthropic’s Claude Code to move laterally through a Taiwanese company’s systems. That’s notable — Claude Code is a developer tool, not a chatbot, and its agentic capabilities make it useful for the kind of multi-step system navigation that previously required a human operator.
A UK AI Safety Institute study adds context: open models have seen their cyber capabilities jump sharply, though they still trail Western frontier models like Claude Mythos by several months for fully autonomous attacks. The gap is narrowing, and TeamT5’s data suggests attackers aren’t waiting for parity. They’re using what’s available now, and the volume increase speaks for itself.
