Google Warns Fully Autonomous AI Cyberattacks Likely Within Three to Six Months
Google's Threat Intelligence Group and Mandiant project that fully autonomous AI cyberattacks will emerge within three to six months. The warning follows the discovery of 'exploit_agent,' an autonomous AI tool found on a server with Russian-language comments that finds and exploits software flaws without human intervention. Zero-day exploits tracked by Google averaged 11 per month in 2026 before surging to 22 in August alone, while vulnerability disclosures nearly doubled from January to August. Google is responding by marketing its AI Threat Defense combining Gemini, Mandiant, and patching agent CodeMender.