State-linked hacking crews more than doubled their attack volume once they started handing the boring parts to a model
The Taiwanese research firm TeamT5 told Bloomberg on Monday that state-affiliated cyber groups have more than doubled the number of attacks they run since they began delegating routine work to AI models and using them to write malicious software. DeepSeek's open models are the favourite, the firm says, because they are cheap, easy to customise and carry light safety guardrails. One documented case shows the shape of it: a Chinese-speaking operator wired DeepSeek into the open-source Hermes Agent framework and, from a single instruction sent over Telegram, had it hunt for internet-facing systems, pick public exploits and try them against more than 460 targets.
Why it mattersThe worry about AI and offensive security has mostly been framed as capability — will a model find a novel exploit a human could not? This finding is about throughput instead, and it is the more immediate problem. The same crew, the same skills, the same known exploits, simply run at twice the rate because the tedious reconnaissance no longer costs human hours. Defence budgets and staffing were sized against the old rate.
Read in the app — free, in 9 languages
Related stories
The models that broke out of their own test to cheat now have a state prosecutor asking for the logs2026-08-25A question mathematicians have not been able to answer since 1948 was answered on Sunday, in a 108-page file posted to X
2026-08-25By the end of last year, nine in ten biomedical papers carried the fingerprints of a language model
2026-08-24Perplexity's revenue tripled in eight months. Nvidia is now talking about buying a piece of it at $30 billion.
2026-08-24Developers noticed the effort dial reading low. Anthropic says it is testing serving configs, and the number means something different this week.
2026-08-24