Hacker used AI agents and ~$5,000 in tokens to breach 100+ corporate sites
deanwball · x · 2026-10-19
Reuters reports a Chinese-speaking hacker used the open-source Cairn agent framework (pairing DeepSeek models with a Claude Code harness) to compromise websites of 100+ companies including Deere, SharkNinja and Cummins, after attempting intrusions on over 1,200 sites.
- Total cost: roughly $5,000 in AI tokens — about $50 per successful breach
- Successful intrusions took 23 hours on average; one campaign's agent instructions fit in fewer than 100 Chinese characters
- Researcher Snorre Fagerland found two dozen maliciously used Cairn instances exposed online, letting him observe the attacks and confirm that commercially available AI agents can execute complex intrusions end-to-end without human intervention
- Victims say they've remediated with no evidence of sensitive data access; whether data was extracted remains unclear
His verdict: "It's a new world now."
More from AGI Musings
- Poisoned SERPs hit 25% of tracked keywords; AccuRanker explains Google's anti-scraping purge — lilyraynyc · 2026-10-19
- Amateurs Are Using GPT-6 to Crack Decades-Old Ciphers, Overwhelming Cryptographers — ChuckDBrooks · 2026-10-19
- From Praying for Miracles to Engineering Them: Why the 2020s Are Unbelievably Lucky — Dr_Singularity · 2026-10-19
- Researcher's cybersecurity collapse prediction confirmed: open models accelerate the breach — davidmanheim · 2026-10-19
- LLM-found zero-days mean most companies' only defense is active security monitoring — davidmanheim · 2026-10-19
- AI agents log 6x human hours at Anthropic as junior hires fall from 27% to 9% — korymath · 2026-10-19