Live updates
Google says 'the era of autonomous attacks has arrived': hackers ran AI that turns patches into exploits with no human in the loop, and a suspected North Korean group backdoored Claude Code settings
First seen on X 20 hours ago@NewsFromGoogle ♥ 2,80041 views
Google says it has, for the first time, caught attackers using AI to fully automate the path from a security patch to a working attack, with no human in the loop. "The era of autonomous attacks has arrived," the company warned.
In its quarterly AI Threat Tracker report released Nov 16, Google Threat Intelligence Group (GTIG) and Mandiant said they found an exposed staging server running a multi-agent framework dubbed exploit_agent. It continuously diffed popular web software before and after patches, pinpointed the fixed flaw, drafted an exploit and executed it automatically. A config file carried Russian-language comments telling the operator to paste API keys "right here, not in the chat"; targets included French media, entertainment and education sites. Disclosed and exploited vulnerabilities nearly doubled from January to August 2026, and zero-day exploits jumped from 11 a month to 22 in August. A suspected North Korea-linked cluster posed as recruiters to breach crypto firms in North America and Europe, then edited .claude/settings.json in internal repos so that launching Claude Code downloaded a backdoor. In "LLMjacking" cases, where stolen cloud credentials are used to run AI models, fraudulent charges topped $100,000 in a single day. "For the most part, attackers aren't using new groundbreaking methods," said GTIG chief analyst John Hultquist; they hit known, unfixed issues "with much greater speed and scale." Google pitched its Gemini-based Google AI Threat Defense alongside the report, drawing replies on X like "create the problem, sell the solution."
Comments 0
Select a sentence in the story (long-press on phones) to comment on it.
“Hackers are now using LLMs to continuously scan software updates, spot newly fixed bugs, and build attacks before teams can apply the fix — all without human involvement.”
