Google warns of autonomous AI attacks finding flaws
Mon, 16th Nov 2026 (Today)
Google Threat Intelligence Group has reported its first observation of a threat actor using an AI agent to identify vulnerabilities and develop exploits without human involvement. The finding marks what it described as a significant threshold in AI-enabled cyber attacks.
Researchers said attackers have moved beyond using large language models for simple prompting and are now deploying autonomous, self-directed attack frameworks that can carry out multiple stages of an intrusion. Google Threat Intelligence Group and Mandiant said this shift has pushed the adversarial AI landscape to a critical inflection point.
According to the report, some newer frameworks can automate vulnerability discovery, draft exploits and execute them with no human in the loop. In one cited case, attackers exploiting newly disclosed vulnerabilities moved from an initial foothold to full domain administrator control within minutes.
Another case involved a framework found on an exposed server that continuously analysed web applications, prepared exploits and ran them automatically. Such systems are compressing attack timelines that once took days or weeks into minutes, the report said.
Rising tempo
The data also pointed to a broader rise in activity around disclosed flaws and exploitation. Vulnerability disclosures and exploits nearly doubled between January and August 2026, while zero-day exploits rose from an average of eight a month in 2025 to 11 a month in 2026 before jumping to 22 in August.
AI-assisted abuse is also surfacing in fraud and influence operations, according to the report. Fraudulent use of stolen access to AI services, sometimes referred to as LLMJacking, generated single-day charges of more than USD $100,000 and, in the most severe cases, reached USD $1 million before being stopped.
Researchers also identified a commercial platform selling AI-enabled influence services that was built to manage more than 100,000 accounts. The report said AI use extended beyond cyber crime into espionage and supply chain attacks, including efforts to tamper with AI coding assistant configuration files to insert backdoors.
Speed over stealth
Google Threat Intelligence Group said many current autonomous attack frameworks still leave clear forensic traces, suggesting they prioritise speed over concealment. That could make them more attractive to cyber criminals willing to accept a higher risk of detection in exchange for faster results.
The assessment drew a distinction between different threat actors. It said criminals and other operators less concerned about remaining hidden are more likely to adopt these tools, while more cautious, stealth-focused actors may be slower to rely on them.
The findings also suggest defenders face a narrowing window to respond once a vulnerability becomes public. As the time between disclosure and exploitation shrinks, patching cycles and internal prioritisation are likely to come under greater pressure.
One issue highlighted in the report is the growing importance of so-called n-day vulnerabilities, flaws that become known after a patch or fix is issued. Automated systems can use those patches as a starting point to identify what changed and build working exploits more quickly.
John Hultquist, Chief Analyst, Google Threat Intelligence Group, said the shift in attacker behaviour is changing the balance of risk. "Early on, much of the conversation around threat actor use of LLMs centered around finding brand-new, unknown flaws, or zero-days. 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. In our latest GTIG AI Threat Tracker, we highlight a case we found of an agentic framework reversing security patches autonomously without human intervention. This underscores the growing n-day problem - it's far easier to reverse-engineer a patch using this technology than it is to hunt for a zero-day. Enterprises are going to be under enormous pressure to speed up their patch cycles in the near future, and may even have to make some hard choices to keep up," Hultquist said.
Defensive steps
The report set out several practical measures for organisations. These included shortening patch timelines, checking whether AI-generated exploits are functional, protecting AI application programming interface keys and cloud usage quotas, and securing configuration files used by AI coding assistants and agents.
It also pointed to the risk that AI-generated malicious code or exploit drafts may still fail in practice, meaning security teams should verify claims of functionality rather than assume every AI-produced output is immediately usable. Even so, the report argued that the overall direction is clear as automated systems become more active in the attack chain.
Google said its response includes model-level safeguards and internal security tools aimed at vulnerability discovery and hardening against machine-speed attacks. The report's central finding, however, was that autonomous AI use by attackers is no longer limited to experimentation and has now been observed in active exploitation.