🔍 Read the full analysis: Did AI Attack Without A Human In The Loop? What Google Says Is Known on ThorstenMeyerAI.com
Get hardware and tech essentials delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Google Threat Intelligence Group says it found an exposed server running a framework that automatically analyzed software patches, developed exploits and executed them without a person approving each step. Google describes this as one observed case; its forecast of attacks with little or no human interaction in coming months is a projection, not a report of widespread fully autonomous attacks.
Google’s Threat Intelligence Group says it found an exposed server running a framework that analyzed software updates, inferred which security flaws they fixed, wrote exploits and executed them without a person approving each step. The finding documents at least one case of automated vulnerability discovery and exploitation, while Google’s prediction that attacks will soon operate with little or no human interaction remains a forecast.
The finding appears in a November 16 report by Google Threat Intelligence Group, produced with Mandiant. Google said the server hosted a multi-agent framework called exploit_agent, connected to commercial and locally run large language models. Its agents compared software code before and after security updates, attempted to identify the vulnerabilities those updates addressed, then generated and ran attacks, according to the report.
Google also said the server contained victim data and scripts for later activity, including examining networks and installing web backdoors. Data on the server suggested the operator targeted French websites in media, entertainment and education. Google did not identify the server’s operator or say how many sites were compromised. A figure in its report shows the framework inspecting access checks in a WordPress plugin; the report does not establish that this examination resulted in an attack.
Google called this the first time it had seen attackers use large language model frameworks to automate vulnerability discovery and exploitation without a human in the loop. That description refers to the observed framework and its workflow, not proof that every stage of a broader attack was autonomous or that such attacks are already common.
Google threat report · Nov 16, 2026
How far AI attacks have come, by Google’s count
Google sorts attacker use of AI into seven stages. Four are history, one is now, two are a forecast.
Google’s seven stages
Observed · end of 2025 – Sep 2026
- 1Prompting a chatbothuman-driven
- 2AI built into malwareautomated runtime call
- 3AI automation in operationsAI in the loop, humans in charge
- 4Experimenting with autonomous frameworksexperimental autonomy
Now · in the wild
- 5AI tools run several attack stages“minimal human-in-the-loop”
Projected · forecast, not observed
- 6Simple attacks with little to no human interactionnear-term, 1–3 months
- 7Sophisticated, fully autonomous attacksmedium-term, 3–6 months
Two headlines, one report
“Have arrived”
The Decoder, Nov 16
Follows one observed case: a framework that found and exploited flaws with no human in the loop.
Evidence · one server“Coming soon”
BleepingComputer, Nov 16
Follows the timeline: stages 6 and 7 are Google’s forecast.
ForecastSolid: observed by Google. Dashed: projected by Google.
Sources: Google Threat Intelligence Group, “GTIG AI Threat Tracker”, Nov 16, 2026 (Figure 2, LLM maturity timeline); GTIG and Mandiant, Oct 30, 2026; BleepingComputer and The Decoder, Nov 16, 2026. Stage names shortened by us.
Graphic: Powered by Thorsten Meyer AI
Automation Moves Into Exploitation
The report matters because it describes AI tools doing more than generating text, writing code on request or assisting a human operator. In Google’s account, the framework linked several steps—finding a possible vulnerability, creating an exploit and running it—without a person signing off on each action. That could let an attacker test software and act at a speed that is difficult for defenders to match manually.
But a single observed framework is not evidence that autonomous attacks are now routine. Google’s own timeline marks the current state as “minimal human-in-the-loop” and places attacks with little or no human interaction in future stages. The distinction matters for organizations weighing their immediate exposure: the report supports taking automated exploitation seriously, but does not establish how often exploit_agent was used successfully or how widely it has spread.
Google also says current tools trade stealth for velocity. Fast, noisy activity may produce signs that defenders can detect, although the report does not quantify detection rates or describe outcomes for particular victims. Google recommends prioritizing actively exploited software flaws, shortening the life of passwords and keys, restricting administrative access and watching for unusually rapid bursts of activity. These are defensive measures, not a guarantee against compromise.
Products Worth Considering

LG 27GS60QC-B Ultragear 27-inch Curved Gaming Monitor (QHD, 180Hz, 1ms, 1000R)

Samsung 27-Inch Odyssey G5 QHD Gaming Monitor (G50F)

Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite
As an affiliate, we earn on qualifying purchases.
Google’s Seven-Stage Threat Timeline
The report, titled GTIG AI Threat Tracker: Agentic Threat Activity Crosses a Significant Threshold, is Google’s quarterly review of AI use by cybercriminals, state-linked groups and influence operations. Google said its findings draw on Mandiant incident-response work, threat tracking and platform defenses. It presented exploit_agent as its central evidence for a threshold in automated exploitation.
Google’s seven-stage timeline separates observed activity from projections. It places the current state at stage five, “minimal human-in-the-loop”, and labels later stages as projected. The company expects attacks with little or no human interaction in the near term, which it defines as one to three months, and fully autonomous attacks in the medium term, three to six months. Those time ranges are Google’s expectations, not confirmed deadlines.
The report also describes other AI-enabled tools that Google says have appeared in real intrusions. It names frameworks including CAIRN and PentAGI, and says it observed CAIRN used against targets that included the U.S. aerospace sector and a customized PentAGI platform used against organizations in Asia. Those examples show use of AI frameworks in intrusion activity, but they do not establish that those operations were fully autonomous.
“In the past quarter, GTIG identified threat actors using LLM frameworks to fully automate vulnerability discovery and exploitation without a human in the loop for the first time.”
— Google Threat Intelligence Group
Scope and Operator Remain Unknown
Google did not name the person or group behind exploit_agent, say how long the server had been operating, or report how many websites were compromised. Its account of interest in French media, entertainment and education websites is based on data found on the server; it does not provide a confirmed victim count. The WordPress plugin example also does not establish that the inspection led to exploitation.
The report does not show whether the framework independently selected its targets, whether a human set its broader objectives or constraints, or how often its exploits succeeded. Google’s claim concerns the absence of human approval at each step in the described workflow. That is narrower than proof that the full operation—from target selection through follow-on activity—ran without people.
It is also unclear how many other operators have comparable capabilities. Google’s timeline forecasts more autonomy within months, but its report does not give a measurable estimate of how frequently such attacks will occur or establish that its projected schedule will hold.
Defenders Watch for Faster Intrusions
Google’s immediate defensive guidance is to patch actively exploited vulnerabilities first, limit the duration and reach of credentials, restrict administrative access and monitor for activity bursts inconsistent with ordinary human operation. Organizations will need to assess these steps against their own systems and threat exposure; the report does not prescribe a single control that prevents this kind of attack.
The next milestones are whether investigators identify the exploit_agent operator or victims, and whether Google or other security teams document comparable cases. Google’s forecast points to a possible increase in attacks requiring little human interaction over the coming months, but the evidence available now is one reported framework plus other AI-assisted tools seen in intrusions. Further reporting will be needed to determine whether the technology becomes more common, more capable or harder to detect.
Key Questions
Did Google confirm that AI carried out an attack entirely on its own?
Google said it found a framework that automatically identified possible vulnerabilities, generated exploits and executed them without human approval at each step. The report does not establish that every part of the operation, including target selection, was done without a person.
What is exploit_agent?
Google describes it as a multi-agent vulnerability research and weaponization framework hosted on an exposed server. It reportedly used commercial and locally run AI models to compare software versions, infer flaws fixed by updates, create exploits and run them.
How many victims were identified?
Google did not provide a victim count. It said data on the server suggested interest in French websites in media, entertainment and education, but did not identify the operator or confirm how many sites were compromised.
Does Google say fully autonomous attacks are already common?
No. Google says it observed a framework that automated vulnerability discovery and exploitation, while describing the current overall stage as having minimal human involvement. More fully autonomous attacks appear in its projected future stages.
What does Google recommend organizations do?
Google recommends prioritizing patches for actively exploited flaws, limiting long-lived passwords and keys, restricting administrative access and monitoring for unusually rapid bursts of activity. These measures can reduce exposure but do not guarantee protection.
Source: ThorstenMeyerAI.com
