Google Threat Intelligence Group says it documented the first case of a threat using a large language model framework to find a flaw and build a working exploit, with no human involved.In its quarterly AI Threat Tracker, published on November 16, the team announced:“The era of autonomous attacks has arrived."The ominous warning continued:"Defending against machine-speed threats at human speed isn’t feasible.”According to Google, the framework was found on an exposed staging server and connects to multiple frontier and locally hosted models. It compares web application code before and after a patch announcement, works out which flaw was closed, drafts an exploit, and runs it automatically.Here's what Google’s report foundThe server also held data taken from victims and scripts for reconnaissance, brute-forcing and installing webshells, with configuration comments written in Russian. Victim telemetry pointed to French media, entertainment and education websites. Google did not name a group or a country.The automation targets n-days, flaws that already have a fix, which shrinks the gap between a vendor shipping a patch and a customer installing it. Vulnerability disclosures and exploits nearly doubled between January and August 2026, and monthly zero-day exploitation rose from an average of eight in 2025 to 11 this year before reaching 22 in August. GTIG also saw the open-source framework CAIRN used against US aerospace targets and PentAGI against organizations across Asia.Google projected that simple attacks run by multiple AI agents with almost no human involvement could appear within three months. Sophisticated fully autonomous attacks are likely within six months.In a statement, John Hultquist, the Chief Analyst at Google Threat Intelligence Group, said:“The reality we're facing today is threat actors are using this tech to operate at speed and scale, and they don't need zero-days to do it... Enterprises are going to be under enormous pressure to speed up their patch cycles.”The 178 companies hit were a different attackOn the same day, web security firm Report URI disclosed a supply-chain attack that reached visitors of 178 organizations through Optimizely Web, the A/B testing tool brands embed in their pages. The list includes HSBC, Equifax, Air France-KLM, Zoom, the BBC, CNN, Salesforce, GitLab, IKEA, Ledger and Peacock. Its founder, Scott Helme, found 188 infected snippets among 1,611 he examined.The script swapped crypto wallet addresses for the attackers' addresses, skimmed card details typed into some site-owned forms, collected cookies and screenshots, and pushed iOS exploit kits at older iPhones. Report URI verified about $7,040 in crypto reaching attacker wallets by 19:25 UTC on November 16.While neither Google nor Optimizely has tied the 178-company Optimizely supply-chain attack to an autonomous AI exploitation framework, both topics are circulating in current cybersecurity news. Optimizely said an unauthorized party exploited a flaw in its account registration process to add code to customer projects, then removed it, revoked tokens, and reset access.For the latest scoops on TV shows and movies, follow SoapCentral.