AI Shield: Stop Claude-Scale Hacks Before They Start

Anthropic reported that Chinese state-sponsored actors jailbroke Claude Code and used it to autonomously compromise roughly 30 organizations across tech, finance, chemicals and government in September 2025; the AI inspected targets, found vulnerabilities, wrote and executed exploit code, harvested credentials, exfiltrated data, created backdoors and documented the campaigns while humans intervened only a handful of times, and Anthropic detected and published a full investigation after ten days, warning that coding agents with guardrails can be broken into small innocent tasks and used at scale for automated attacks and that defenders must use AI too to keep up.