🔍 Read the full analysis: Google’s Findings On AI Attacks: What Is Demonstrated And What Is Projected on ThorstenMeyerAI.com
Get the latest gadgets 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. The finding is a reported example, not evidence that fully autonomous attacks are widespread; Google projects attacks requiring little human input within one to three months.
Google Threat Intelligence Group said it found an exposed server running exploit_agent, a framework that compared software before and after security updates, inferred the flaws they fixed, and then wrote and executed exploits automatically. Google called it the first case it had seen of attackers fully automating vulnerability discovery and exploitation without a person in the loop, but said more autonomous attacks remain a forecast rather than an established norm.
The finding appears in GTIG’s AI Threat Tracker, published November 16 with Mandiant. Google said its report draws on incident-response work, tracking of threat actors and defenses on its platforms. It described exploit_agent as a multi-agent system connected to several commercial and locally run large language models. Its reported workflow included monitoring widely used software, comparing code before and after patches, identifying possible vulnerabilities, drafting attacks and executing them.
Google also said the server held stolen victim data and scripts for later activity, including exploring networks and planting web backdoors. Data found there suggested an interest in French websites in media, entertainment and education. Google did not report how many sites were compromised, identify the server’s operator or establish that every vulnerability the framework examined was exploited.
The report describes other AI-assisted activity in real intrusions. Google named frameworks including CAIRN and PentAGI, saying it saw CAIRN used against targets including the U.S. aerospace sector and a customized PentAGI platform used against organizations in Asia. These observations concern the use of frameworks in attacks; they do not show that those operations were fully autonomous from start to finish.
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 Is Real, Scale Is Unknown
The report matters because it describes automation applied to a consequential part of cyberattacks: turning a newly released patch into a possible route of exploitation. If a system can identify what a fix changed and act on that analysis quickly, organizations may have less time to update exposed software before attackers test it. Google’s account points to a specific observed framework, however, not evidence that this capability is common across threat actors or routinely successful.
Google’s own maturity scale distinguishes the finding from the larger risk it expects. The report places activity at stage 5 of seven, characterized as involving minimal human input. It projects attacks with little to no human interaction in the near term, defined as one to three months, and fully autonomous attacks in three to six months. Those are Google’s forecasts, not confirmed events or guaranteed timelines.
The company also says current tools trade stealth for speed: rapid, automated activity can be noisy and easier to spot. For defenders, that means the reported capability raises the value of prompt patching and monitoring, while leaving open how often these systems succeed and whether their activity can remain concealed.
Popular right now

Vitamix Nano Personal Blender with SmoothiePilot Technology, Black

Ninja Creami Scoop & Swirl Ice Cream Maker – Bone & Champagne (Amazon Exclusive)

Dyson V9 Motorbar Cordless Vacuum
As an affiliate, we earn on qualifying purchases.
From One Server to a Forecast
GTIG presents the report as a quarterly review of how criminal groups, state-backed actors and influence operations use large language models. In an October 30 post, Google outlined a seven-stage progression toward autonomous attacks. At the later stages, an operator would set a target and objective while agents carry out more of the attack chain; the final stage adds greater stealth and sophistication.
The November report labels stage 5 as the current stage and says it is observed in the wild. Stages 6 and 7 are projections. This distinction helps explain why coverage of the same report can describe automated attacks as already present or as coming soon: the first description refers to exploit_agent’s reported automation of vulnerability discovery and exploitation, while the second refers to broader attacks with little or no human involvement.
Google’s evidence for the exploit_agent claim centers on one exposed staging server. That is meaningful evidence of a capability in at least one case, but it does not establish how many operators use similar systems or how widespread the practice is.
“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
How Far the Activity Reached
Several central details remain unknown. Google did not say who operated exploit_agent, how many websites it accessed or compromised, or whether the server’s analysis of a particular software patch led to a successful attack. The data indicated possible interest in French sites across three sectors, but does not establish the full extent or outcome of targeting.
The report also does not show how widespread this level of automation is. One observed framework cannot establish that fully automated exploitation is now routine, and Google’s one-to-three-month and three-to-six-month forecasts are projections, not confirmed milestones. The available account does not quantify how often the tools succeed or how often defenders detect them.
Defenses and the Next Milestones
Google’s report advises organizations to prioritize patches for flaws known to be actively exploited, reduce the use of long-lived passwords and keys, restrict administrative access and watch for unusual bursts of activity. Those steps address the possibility that automated systems can move quickly from vulnerability research to exploitation; they do not depend on the broader forecast coming true.
The next test will be whether Google or other researchers report additional cases showing attacks carried out with little human involvement, and whether the projected stages appear within the stated windows. Until then, the distinction remains: Google says one fully automated vulnerability-exploitation case was observed, while widespread or fully autonomous attacks remain unconfirmed.
Key Questions
What did Google say it found?
Google said it found an exposed server hosting exploit_agent, a framework that analyzed software changes around patches, drafted exploits and executed them automatically.
Does the report show fully autonomous cyberattacks are widespread?
No. Google reported a specific observed framework and placed current activity at stage 5, described as involving minimal human input. It did not establish that fully autonomous attacks are widespread.
What does Google forecast?
Google projects attacks with little to no human interaction within one to three months, and fully autonomous attacks within three to six months. These are forecasts, not confirmed developments.
Who did the server target?
Google said data on the server suggested interest in French websites in media, entertainment and education. It did not disclose the number of sites compromised or identify the operator.
What should organizations do now?
Google recommends prioritizing patches for actively exploited flaws, limiting long-lived credentials, restricting administrative access and monitoring for unusually rapid bursts of activity.
Source: ThorstenMeyerAI.com
