🔍 Read the full analysis: What’s Known About AI Attacks Without Human Involvement, According To Google on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Google Threat Intelligence Group says it found one exposed server running software that used AI agents to identify vulnerabilities and execute exploits without a person approving each step. Google describes this as a first observed case, but says attacks that operate with little or no human interaction more broadly are still projected, not established as routine.
Google Threat Intelligence Group said it identified an exposed server running a multi-agent system that analyzed software updates, inferred the vulnerabilities they fixed, and automatically executed attacks without a person approving each step. The finding, described in a report published November 16 with Mandiant, is Google’s first observed instance of attackers fully automating vulnerability discovery and exploitation; it does not establish that fully autonomous attacks are already widespread.
Google said the server hosted a framework it dubbed exploit_agent, connected to multiple commercial and locally run language models. The system reportedly monitored popular web software, compared code before and after security patches, inferred what flaw a patch addressed, drafted an exploit, and ran it. Google described the discovery in its report, “GTIG AI Threat Tracker: Agentic Threat Activity Crosses a Significant Threshold.”
The company also said the server contained stolen victim data and scripts for possible follow-on activity, including searching within networks and installing web backdoors. Data on the server suggested an interest in French websites in media, entertainment and education. Google did not identify the operator or state how many sites were compromised. Its report also illustrates the framework examining a WordPress plugin’s access checks, but does not establish that this particular analysis resulted in an attack.
Google’s finding concerns a system observed on one exposed server, while other parts of the report describe AI-assisted tools used in intrusions. The company says attackers have used frameworks including CAIRN and customized PentAGI in real operations. Those examples do not, by themselves, show that every stage of an attack 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
Why Autonomous Exploits Matter
Automating the move from a software patch to a working exploit could let attackers test and act on newly disclosed weaknesses faster than a human-led process. For defenders, that creates pressure to prioritize patching flaws already known to be exploited and to spot unusual activity quickly. Google’s report points to a possible change in the speed of vulnerability exploitation, rather than evidence that attackers have already replaced people across the full attack chain.
There is also a practical limit in Google’s account: the company says current tools “trade stealth for velocity.” Fast, automated activity may generate patterns that security teams can detect. That is not a guarantee of detection, and the report gives no measured success rate for the system. The balance between speed and concealment will matter as defenders evaluate both the threat and the effectiveness of their monitoring.
Google recommends focusing on actively exploited vulnerabilities, reducing long-lived passwords and access keys, restricting administrative access, and monitoring for bursts of activity inconsistent with ordinary human behavior. These are defensive steps tied to the reported risk; the report does not claim that any one measure would prevent every AI-assisted intrusion.
Products Worth Considering

Bose Smart Dolby Atmos Soundbar with Alexa, A.I. Dialogue Mode, and Bluetooth, Black

Amazon Kindle Scribe (16GB) with Premium Pen, Tungsten

LG 27GR83Q-B 27-inch Ultragear QHD IPS Gaming Monitor
As an affiliate, we earn on qualifying purchases.
Google’s Seven-Stage Threat Timeline
Google’s report places current activity at stage 5 of a seven-stage timeline, labeled “Minimal human-in-the-loop” and “in the wild.” In Google’s description, attackers still have some role, even as frameworks automate parts of their operations. The company projects attacks with little to 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.
That distinction explains why accounts of the report can sound different. The observed exploit_agent system supports the statement that fully automated vulnerability discovery and exploitation has occurred in at least one case. The later stages on Google’s chart are projections, not observations that those capabilities are already common. Google has described a later stage as one where a person sets the target and objective while a multi-agent system carries out the attack chain; a further stage would add greater stealth and sophistication.
The threat tracker draws on data Google says comes from Mandiant incident-response work, global tracking of threat actors, and live platform defenses. It is a company threat-intelligence assessment, not a peer-reviewed study, and the report’s account of exploit_agent rests on a specific server discovery rather than a broad prevalence estimate.
“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
What the Server Evidence Cannot Show
Google has not publicly identified who operated exploit_agent, how many organizations were affected, or whether the server’s stored data corresponds to successful compromises. The company’s report also does not establish whether every attack step, from choosing targets through maintaining access, was automated. Its “first time” claim is specifically about fully automating vulnerability discovery and exploitation without a human in the loop.
The report does not say whether the WordPress plugin analysis shown in its illustration led to exploitation, or provide a success rate for the framework. It also offers no estimate of how often comparable systems are being used. Google’s forecast for broader autonomy over the next one to six months should be read as a prediction, not confirmation that such attacks will become routine on that schedule.
What Defenders Should Watch
Google’s report sets no public deadline for a further announcement about exploit_agent. Its near-term outlook instead points security teams toward preparing for attacks that may need less direct human involvement. Organizations can review patch priorities, tighten administrative access, reduce the lifetime of credentials, and monitor for rapid sequences of activity that appear automated.
The next meaningful evidence will be whether Google or other incident responders document additional cases, describe the extent of any victim impact, or report that attackers are using the tools across a broader range of operations. Until then, the distinction remains: one automated exploitation system was observed, while more fully autonomous attacks at scale remain a forecast.
Key Questions
What did Google say it found?
Google said it found an exposed server hosting exploit_agent, a multi-agent framework that analyzed software patches, inferred vulnerabilities, drafted exploits, and automatically executed them.
Does Google say fully autonomous cyberattacks are now common?
No. Google reports one case of automated vulnerability discovery and exploitation. It labels the current stage “Minimal human-in-the-loop” and treats broader, fully autonomous attacks as a future projection.
Who was targeted by the system?
Google said data on the server suggested interest in French websites in media, entertainment and education. It did not name the operator or report how many sites were compromised.
When does Google expect more autonomous attacks?
Google’s timeline projects attacks with little to no human interaction within one to three months and fully autonomous attacks within three to six months. Those are forecasts, not confirmed milestones.
What steps does Google recommend for defenders?
Google advises prioritizing actively exploited flaws, limiting long-lived passwords and access keys, restricting administrative access, and watching for unusually rapid bursts of activity.
Source: ThorstenMeyerAI.com
