🔍 Read the full analysis: Google Flags AI Attacks Without Human Control: Fact Versus Forecast on ThorstenMeyerAI.com
Get the little things that make your day 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’s Threat Intelligence Group says it found an exposed server running a multi-agent framework that could identify vulnerabilities from software patches, generate exploits and execute them without a person approving each step. Google describes this as one observed case; its prediction that attacks with little or no human interaction will become more common in the coming months is a forecast, not a report of widespread autonomous attacks.
Google’s Threat Intelligence Group says it found an exposed server running a framework that identified software vulnerabilities, wrote exploits and launched them without human approval at each step—a development that shows automated exploitation has occurred, but does not establish that fully autonomous attacks are widespread. In a report published November 16 with Mandiant, Google described the server as one observed case and forecast that attacks requiring little or no human interaction could emerge more broadly in the coming months.
The report, titled GTIG AI Threat Tracker: Agentic Threat Activity Crosses a Significant Threshold, says the framework, named exploit_agent, was hosted on an exposed staging server. Google says it connected multiple AI agents to commercial and locally run large language models. The system compared software code before and after an update, inferred what vulnerability the patch addressed, drafted an exploit and automatically executed it.
Google also found stolen victim data and scripts for later activity on the server, including steps for examining a network and installing web backdoors. Data on the server suggested the operator focused on French websites in media, entertainment and education. Google did not identify who operated it or say how many sites were compromised. A report image shows the framework examining access checks in a WordPress plugin, but Google did not say whether that analysis led to an attack.
Google said it drew its findings from Mandiant incident-response work, global tracking of threat actors and defenses operating on its platforms. Beyond exploit_agent, the report describes other AI-assisted activity, including use of open-source testing frameworks in intrusions. Google cited CAIRN activity targeting organizations that included the U.S. aerospace sector and a customized PentAGI platform used against organizations in Asia.
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 Up Exploit Work
The finding matters because the framework reportedly joined several stages of an attack that are often handled separately: spotting a vulnerability, turning it into working attack code and running that code. Automating those tasks could let an operator move faster, particularly after a software maker releases a patch that reveals what needs fixing.
That does not mean the tools can reliably carry out every part of an intrusion on their own. Google describes the current state as “minimal human-in-the-loop” and says the tools trade stealth for speed. Fast, high-volume activity may create more opportunities for defenders to detect it. The report therefore points both to a greater need for rapid patching and to limits in what the observed tools can do discreetly.
For organizations, Google’s practical advice is to prioritize flaws known to be actively exploited, reduce the use of long-lived passwords and keys, restrict administrator access, and monitor for bursts of activity that are unlikely to come from a person. These measures address the speed and access risks raised by the report without assuming that every organization is facing fully autonomous attacks.
Products Worth Considering

Ring Alarm 14-Piece Home Security Kit

LG OLED evo C4 55-Inch 4K Smart OLED TV (OLED55C4PUA) with 144Hz Refresh Rate, HDMI Cable, 24-Month Protection Plan, and Streaming Bundle

Blueair Blue Pure 211i Max Smart Air Purifier for Large Rooms
As an affiliate, we earn on qualifying purchases.
Observed Activity Versus Forecast
Google’s report separates what it says has happened from what it expects next. It calls exploit_agent the first instance it identified of threat actors using large language model frameworks to fully automate vulnerability discovery and exploitation without a human in the loop. That is a claim about Google’s observed evidence; the report’s account centers on one exposed server, not a measure of how common the technique is.
Separately, Google’s seven-stage maturity chart places current activity at stage five, described as “minimal human-in-the-loop”. It projects attacks with little to no human interaction in one to three months and fully autonomous attacks in three to six months. Those timeframes are Google’s forecasts, not confirmed milestones. Its October 30 explanation 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. Coverage saying automated attacks have arrived refers to Google’s exploit_agent observation; coverage saying they are coming soon refers to the broader projected stages. The evidence supports the narrower claim that Google found one framework automating exploitation, while the wider expectation remains prospective.
““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 Of The Server Finding
Several details remain unknown. Google did not identify the server’s operator, report how many websites were successfully breached, or establish whether the WordPress plugin analysis shown in its report resulted in exploitation. The data indicated an apparent focus on French media, entertainment and education sites, but the report does not provide a full account of targets or outcomes.
The evidence also does not show how often exploit_agent or comparable systems are being used across attacks. Google’s timeline for attacks with little or no human involvement is a prediction, not confirmation; the report gives no assurance that the projected stages will arrive on that schedule. The extent to which these systems can conduct stealthy, complex operations without human oversight remains unclear.
Defenses And Further Evidence
Google’s report does not announce a specific enforcement action or a confirmed next incident. The immediate response it recommends for defenders is operational: patch actively exploited flaws first, limit long-lived credentials, secure administrator accounts and watch for unusually rapid activity. Organizations can also review whether their monitoring would surface automated attempts against newly patched software.
Further reporting or incident-response findings may clarify who operated exploit_agent, which systems were affected and whether similar frameworks are appearing in more intrusions. Until then, the clearest distinction is between one reported case of automated vulnerability exploitation and Google’s forecast of broader, more autonomous attacks in the months ahead.
Key Questions
What did Google say it found?
Google said it found an exposed server hosting exploit_agent, a framework that compared software versions around a patch, inferred the vulnerability, drafted an exploit and executed it without human approval at each step.
Does the report show that fully autonomous attacks are widespread?
No. Google described an observed case of automated vulnerability discovery and exploitation, while its prediction of attacks with little or no human interaction is about what may happen in coming months. The report does not establish how common these systems are.
Who or what was targeted?
Data on the server suggested interest in French websites associated with media, entertainment and education. Google did not identify the operator or state how many sites were compromised.
What should organizations do now?
Google advised organizations to prioritize actively exploited software flaws, reduce reliance on long-lived passwords and keys, restrict administrator access and monitor for bursts of activity that appear automated.
Source: ThorstenMeyerAI.com
