Google Threat Intelligence Group (GTIG) reportedly identified an operation using exploit_agent to analyze software patches and launch exploits, according to El Español’s November 18, 2026 account. The reported sequence points to attacks on systems that have not installed a fix for a disclosed flaw.

On November 16, Google Cloud Security posted about a new GTIG AI Threat Tracker describing autonomous patch-to-exploit pipelines. The announcement followed Google’s earlier warning about agentic AI cyberattacks, which NeoTeo covered previously.

How exploit_agent reportedly moved from patch to exploit

El Español described exploit_agent as a multi-agent framework connected to commercial and locally run AI models. The account says it compared software before and after a patch, inferred the flaw the patch addressed, then drafted and launched an exploit. El Español characterized the finding as GTIG’s first observed real-world case of an end-to-end automated vulnerability-discovery and exploitation workflow.

That sequence starts with a patch: the reported approach uses the changes to identify a flaw and directs an exploit toward systems that have not yet received the fix. It is a patch-to-exploit path involving a disclosed vulnerability.

A published patch can reveal the flaw it fixes; systems still awaiting that fix face the risk of a patch-to-exploit attempt.

What the reported activity says about the operation

El Español reported that GTIG found stolen data and tools for internal reconnaissance, password guessing and backdoor installation on the exposed server. Activity records reportedly indicated interest in French websites in media, entertainment and education.

El Español also reported Russian-language comments and said the operators had not been identified.