The gap between a published patch and a working exploit is being automated. Google Threat Intelligence Group (GTIG) says that in the past quarter it identified threat actors using large language model frameworks to fully automate vulnerability discovery and exploitation without a human in the loop for the first time.
In its quarterly AI threat report, published on November 16, GTIG describes an exposed staging server that hosted a multi-agent framework it calls exploit_agent. The framework runs continuous differential analysis of popular web application software. It compares code before and after patch announcements, analyzes the findings to identify potential vulnerabilities, drafts exploits and executes them automatically. GTIG also found data exfiltrated from victims on the server, along with scripts for internal reconnaissance, brute-forcing and installing webshells. Victim telemetry suggested the operator was interested in French media, entertainment and education websites.
GTIG explains why attackers favor patched flaws. Weaponizing a known vulnerability lets them focus on a finite set of material, such as two versions of the source code, the release notes and published research. Its figures show overall disclosures and exploits nearly doubled between January and August 2026, while zero-day exploits rose from a 2025 average of 8 a month to 11 a month, then reached 22 in August.
The report adds a second cost. A lower barrier to entry has produced a wave of LLM-generated exploits that are claimed to work and do not, which leaves defenders to verify or debunk them. GTIG cites the PHP vulnerabilities patched on October 21, where hundreds of non-functional proofs of concept appeared in the first days.
In our reading, the finding changes how a patch service level should be written for internet-facing web applications. The clock for those systems should start when a fix or a preview build becomes public, because that is the material the framework reads. Amazon’s telemetry on PHP2Shell showed an attempt four days before disclosure. On the same day, we reported that Project Zero set out a shorter disclosure schedule.
Source: Google Threat Intelligence Group