Google’s Threat Intelligence Group (GTIG) says it has seen the first fully automated cyber‑attack chain run by an AI agent. The system, named "exploit_agent", watches software updates, compares code before and after a patch, spots a flaw, writes an exploit and runs it – all without a human touching the code. GTIG found the server running the tool in November 2026 and traced its activity to Russian‑language configuration files. The operator appears to have focused on French websites in the media, entertainment and education sectors. Google rates the current activity at Level 5, meaning AI handles most stages of the attack while a human may only set it up. The company warns that more powerful, stealthier attacks could appear within months.
Why this matters now
The shift to autonomous, multi‑agent frameworks changes the speed of attacks. Where a human once had to write and test each exploit, AI can compress the timeline to minutes, letting threat actors strike many targets at once. Google and Mandiant have documented several open‑source tools – ARTEX, CAIRN, CyberStrikeAI, HexStrike AI and PentAGI – that let attackers chain reconnaissance, vulnerability scanning and payload generation automatically. These tools leave forensic traces because they prioritize speed over stealth, which gives defenders a chance to spot the abnormal traffic. However, as underlying language models improve, the trade‑off may shrink, making detection harder.
Broader context
The trend fits a larger pattern Google describes as an “inflection point” for adversarial AI. Over the last quarter, GTIG and Mandiant have seen attackers move from simple prompting of large language models (LLMs) to building entire autonomous pipelines that research, weaponize and monetize vulnerabilities. One new business model, called LLMJacking, steals cloud AI credentials and resells access to high‑performance models on underground markets. State‑linked espionage groups are also using LLMs to gather target data and insert backdoors via compromised coding assistants. While the current automated attacks sacrifice stealth, the report stresses that future versions could hide their footprints, raising the stakes for security teams worldwide.
Why it matters
For ordinary businesses, the news means attackers can now scan for and exploit vulnerabilities at machine speed, increasing the number of potential incidents. Defenders will need to watch for unusual automated traffic patterns rather than rely on human‑driven alerts. It remains unclear how quickly attackers will improve stealth, so organizations should strengthen patch management and monitoring now.