🔍 Read the full analysis: AI Attacks With No Human In The Loop: What Google Knows So Far on ThorstenMeyerAI.com
TL;DR
Google’s threat intelligence team says it found an exposed server running an AI-based framework that identified vulnerabilities from software updates, generated exploits and executed them without a person approving each step. Google describes this as a first observed case, but says attacks operating with little or no human involvement more broadly are still projected for coming months.
Google’s threat intelligence team says it found an exposed server running software that could identify vulnerabilities, write exploits and launch them without human approval at each step—a development that marks the first such case the company says it has observed. The finding, published November 16 with Mandiant, does not establish how many organizations were attacked or who operated the server; Google separately forecasts that attacks with little or no human involvement could become more common in the coming months.
The Google Threat Intelligence Group (GTIG) report describes a multi-agent framework it named exploit_agent. The system was hosted on an exposed staging server and connected to multiple commercial and locally run large language models. According to Google, it monitored software updates, compared code before and after patches to infer what weakness had been fixed, generated an exploit and automatically ran it.
Google also found stolen victim data and scripts for possible follow-on activity on the server, including scripts for exploring networks and installing web backdoors. The data, Google said, suggested the operator was interested in French websites in media, entertainment and education. The report does not identify the operator or say how many sites were compromised. A figure shows the framework examining access checks in a WordPress plugin, but Google does not say whether that examination led to a successful attack.
GTIG said the finding was based on its incident-response work, threat-actor tracking and platform defenses. The report also describes open-source AI security-testing frameworks being used in real intrusions: Google cited CAIRN activity against targets including the U.S. aerospace sector and a customized PentAGI platform used against organizations in Asia. Those examples indicate use of AI-enabled tools in attacks, but do not mean every stage of those operations was autonomous.
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 Changes Attack Speed
Automating the chain from patch analysis to exploit execution could let an attacker act quickly after a software flaw becomes visible through an update. That matters to organizations that delay applying fixes: a vulnerability may be easier to identify once the patch reveals what changed, while automated tools can test and act at a pace that is difficult for people to match.
Google’s report also draws a distinction between a working automated capability and a forecast about widespread autonomy. The company says the observed system traded stealth for speed: such tools may generate noisy activity that defenders can detect. That is a present limitation, not a guarantee of safety. Security teams may need to watch for unusual bursts of scanning or exploitation while continuing to prioritize patching and access controls.
For readers and businesses, the report is a warning about the possible speed of exploitation, not evidence that every attacker now operates an autonomous system. The findings support practical defensive steps, but do not quantify the likelihood that a particular organization will be targeted or compromised.
Popular right now

KitchenAid Artisan 5-Quart Tilt-Head Stand Mixer with Flex Edge Beater

De'Longhi Classic Signature Espresso Machine with Adjustable Milk Frother

TCL F8D Series 50-Inch 4K QLED Smart Fire TV
As an affiliate, we earn on qualifying purchases.
What Google Observed Versus Forecast
GTIG’s November 16 publication, titled “GTIG AI Threat Tracker: Agentic Threat Activity Crosses a Significant Threshold,” is a quarterly assessment of how criminal groups, state-backed actors and influence operations use large language models. Google’s central claim is carefully bounded: in the past quarter, it identified threat actors using LLM frameworks to automate vulnerability discovery and exploitation without a human in the loop for the first time in its observations.
Google’s maturity model places current activity at stage five of seven, labeled “Minimal human-in-the-loop.” It projects attacks with little or no human interaction in the near term—one to three months—and fully autonomous attacks in the medium term—three to six months. Those stages are forecasts, not confirmed descriptions of routine current activity. Google’s October 30 discussion described a later stage in which a person sets the target and objective while multi-agent systems carry out the attack chain.
This distinction helps explain differing headlines about the report. The claim that automated attacks have “arrived” refers to the exploit_agent case; claims that fully autonomous attacks are “coming soon” refer to Google’s projected next stages. The observed example does not establish that autonomous attacks are already the norm.
„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
Who Was Targeted Remains Unknown
Google has not publicly identified the server operator, the specific affected websites or the number of successful intrusions. Its description of French media, entertainment and education sites is based on victim data found on the server and is presented as an indication of the operator’s interests. The report also does not establish whether the WordPress plugin analysis shown in its figure resulted in exploitation.
It is also unclear how often exploit_agent or comparable systems are being used, how reliably they succeed, and whether the activity Google described involved human direction outside the individual discovery-and-exploitation steps. Google’s maturity timeline gives projected windows, not a guarantee that attacks at later stages will appear on schedule. The report says current tools can be fast but noisy; it does not provide a measured detection rate or quantify their effectiveness.
Defenders Face Near-Term Priorities
Google’s report points to immediate defensive measures: prioritize fixes for vulnerabilities known to be actively exploited, reduce reliance on long-lived passwords and access keys, restrict administrative access, and monitor for unusually rapid bursts of scanning or attack activity. These are recommendations arising from the described threat pattern, not a claim that any one measure will prevent every attack.
The next evidence to watch for is whether Google or other researchers report more cases of automated exploitation, provide details about affected systems, or document attacks that progress through a fuller chain with little human input. For now, Google’s report establishes one observed framework and a forecast of further automation; the frequency and practical reach of the next stages remain open questions.
Key Questions
What did Google say it found?
Google said it found an exposed server hosting exploit_agent, a multi-agent framework that could infer vulnerabilities from software patches, generate exploits and execute them without a person approving each step.
Does the report show that fully autonomous attacks are now common?
No. Google describes one observed case of automated vulnerability discovery and exploitation. Its report labels the current overall stage “Minimal human-in-the-loop” and treats more fully autonomous attacks as a forecast.
Who was attacked, and how many sites were affected?
Google has not named the operator or specified the number of compromised sites. It said data on the server suggested interest in French websites in media, entertainment and education.
What does Google predict will happen next?
Google projects attacks with little or no human interaction in the near term, defined as one to three months, and fully autonomous attacks in the medium term, defined as three to six months. These are projections, not confirmed events.
What should organizations do in response?
Google’s recommendations include prioritizing patches for actively exploited flaws, reducing long-lived credentials, restricting administrative access and monitoring for unusually rapid attack activity.
Source: ThorstenMeyerAI.com