🔍 Read the full analysis: What Google’s Report Says About AI Attacks Operating Without Humans on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Google’s November 16 threat report describes an exposed server running a framework that identified software flaws from updates, wrote exploits and executed them without a person approving each step. Google calls this a first observed case, but says attacks operating with little or no human interaction more broadly remain a near-term forecast, not an established norm.
Google’s Threat Intelligence Group said it found an exposed server running software that identified vulnerabilities, drafted exploits and launched them without a person approving each step—a first for the group’s observed cases. The finding, included in a November 16 report with Mandiant, documents one instance of automated exploitation; Google’s forecast that attacks will soon operate with little or no human input describes a possible next stage, not a pattern it says is already widespread.
Google said the server hosted a multi-agent framework it called exploit_agent, connected to several commercial and locally hosted large language models. According to the report, the system monitored widely used web software, compared code from before and after security updates, inferred which weakness a patch addressed, then prepared and executed an exploit. Google characterized this as the first time it had seen threat actors use language-model frameworks to automate both vulnerability discovery and exploitation without a human in the loop.
The server also contained data taken from victims and scripts for possible follow-on activity, including searching within networks and installing web backdoors, Google reported. The data suggested the operator was interested in French websites in media, entertainment and education. Google did not identify the operator or say how many sites were compromised. A configuration file included Russian-language comments, but the report does not establish who wrote them or where the operator was based.
Google’s report also describes the use of open-source AI penetration-testing frameworks in real intrusions. It cited CAIRN activity against targets including the U.S. aerospace sector and a customized PentAGI platform used against organizations in Asia. Those examples are separate from the exploit_agent server finding; the report does not say that every framework it discusses autonomously completed an entire attack.
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 Speeds Exploit Attempts
The finding matters because a system that can infer a vulnerability from a software patch and immediately try an exploit could shorten the time between a fix becoming public and attackers attempting to use the underlying flaw. That may put pressure on organizations to apply security updates quickly, especially for internet-facing software, and to monitor systems for unusual activity after patches are released.
But Google also says the current tools trade stealth for velocity: they can act quickly, while producing noisy activity that defenders may detect. The report does not establish how often this approach succeeds, how many victims it affected, or whether it caused more damage than human-directed attacks. Its evidence supports concern about a new capability, not a claim that autonomous attacks have become routine.
For defenders, the practical steps described in the source are familiar but pressing: prioritize flaws known to be actively exploited, reduce reliance on long-lived passwords and keys, restrict administrative access, and look for bursts or patterns of activity unlikely to come from a person. Those measures address the risk without assuming that every alert reflects an AI-run attack.
Popular right now

KitchenAid Commercial Series 8-Quart Bowl-Lift Stand Mixer, White

TCL 55-Inch F7D Series QLED 4K UHD Smart Fire TV (2026)

Vitamix VX1 Professional-Grade Blender, 64 oz, Black
As an affiliate, we earn on qualifying purchases.
Google’s Seven-Stage Threat Model
Google titled the report “GTIG AI Threat Tracker: Agentic Threat Activity Crosses a Significant Threshold.” It describes the tracker as a quarterly review of AI use by criminal groups, state-backed actors and influence operations. Google said its findings draw on Mandiant incident-response work, global threat tracking and defenses on its platforms.
The report places current activity at stage five of a seven-stage model, labeled “Minimal human-in-the-loop.” Google marks the later stages as projected: it expects 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, three to six months. These are forecasts from the report, not confirmed events on a fixed schedule.
That distinction explains why coverage of the report has used different descriptions. One interpretation emphasizes that Google observed a framework carrying out automated exploitation; another emphasizes that broader, nearly human-free attacks are still projected. Both refer to different claims in the same report. Google’s description of today’s stage is not “no human,” and its evidence for exploit_agent is a specific server discovery.
““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
Victims and Operators Remain Unknown
Google has not disclosed who operated exploit_agent, how many websites were accessed or compromised, or whether the WordPress plugin checks illustrated in the report led to an attack. The data on the server points toward interest in French media, entertainment and education websites, but does not by itself establish the full set of targets or the impact on them.
The report also does not quantify how often AI frameworks successfully exploit vulnerabilities, or show that the observed server represents typical attacker behavior. Google’s one-to-three-month and three-to-six-month timelines are projections; the report does not guarantee that these stages will arrive on schedule or explain how frequently they may occur.
Defenders Face Faster Patch Races
Organizations should follow their normal security response processes while giving priority to actively exploited vulnerabilities, exposed services and privileged accounts. Google’s recommendations in the source include shortening the life of passwords and keys, limiting administrative access and watching for unusually rapid activity. The report does not name a specific new deadline or a particular software product that all organizations must patch.
The next developments to watch are further evidence about real-world use of automated frameworks, details about the exploit_agent operator or victims, and whether incident responders observe attacks matching Google’s projected later stages. Until then, the report supports a clear but bounded conclusion: Google has documented one case of fully automated vulnerability discovery and exploitation, while wider attacks requiring little or no human interaction remain a forecast.
Key Questions
What did Google say the exploit_agent server did?
Google said it compared software code before and after updates, inferred vulnerabilities addressed by patches, drafted exploits and executed them without a person approving each step.
Does Google say fully autonomous cyberattacks are already common?
No. The report describes one observed case of automated vulnerability discovery and exploitation. Google places current activity at “Minimal human-in-the-loop” and treats broader attacks with little or no human interaction as a projection.
Who was targeted by the server’s operator?
Google said data on the server suggested interest in French websites associated with media, entertainment and education. It did not identify the operator or specify how many sites were compromised.
When does Google expect attacks with less human involvement?
Google’s model projects attacks with little to no human interaction in one to three months, and fully autonomous attacks in three to six months. These are forecasts, not confirmed deadlines.
What should organizations do in response?
The source recommends prioritizing actively exploited flaws, limiting long-lived credentials and administrative access, and monitoring for unusually rapid bursts of activity. It does not identify one universal patch or action that applies to every organization.
Source: ThorstenMeyerAI.com
