Read the full analysis: Google’s AI Attack Report: The Facts, The Forecasts, And The Gap on ThorstenMeyerAI.com
TL;DR
Google’s November 16 AI Threat Tracker reports an observed framework, called exploit_agent, that automated vulnerability research and exploitation without a person approving each step. Google describes that finding as distinct from its forecast that attacks with little or no human involvement could emerge more broadly in coming months; the scale and operator of the observed activity are unknown.
Google’s Threat Intelligence Group said it found an exposed server running software that used AI agents to identify vulnerabilities in web applications and automatically execute exploits, without a person approving each step. The finding, published November 16 in a report with Mandiant, documents at least one case of automated vulnerability discovery and exploitation; Google’s forecast that attacks with little or no human involvement will become more common is a separate, forward-looking assessment.
Google named the framework exploit_agent. According to the report, it compared software code before and after security updates, inferred which vulnerabilities the updates addressed, drafted attacks and ran them. The system connected to multiple commercial and locally hosted large language models. Google described the activity as the first time its researchers had seen attackers use LLM frameworks to fully automate vulnerability discovery and exploitation without a human in the loop.
On the same server, researchers found data stolen from victims and scripts for possible follow-on activity, including searching networks and installing web backdoors. Google said victim data suggested interest in French websites in media, entertainment and education. The report also showed the framework examining access checks in a WordPress plugin, but did not establish that this particular examination resulted in a successful intrusion.
The report also describes open-source AI-assisted penetration-testing tools used in real-world intrusions. Google said it observed CAIRN being used against targets that included the U.S. aerospace sector and a customized PentAGI platform used against organizations in Asia. Those examples show attackers using such frameworks in operations, but do not establish that every step of those campaigns 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.
Forecast
Solid: 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 Raises the Pace of Exploitation
The reported capability could shorten the time between a software update and an attacker identifying the vulnerability it fixes. If a system can compare versions, infer a weakness and test an exploit automatically, defenders may have less time to protect exposed systems. That makes prompt patching of actively exploited flaws and close monitoring of exposed services especially relevant.
Google also cautioned that current systems have limitations. The report describes tools that trade stealth for velocity: rapid, automated activity may generate noticeable bursts that defenders can detect. This is not evidence that attackers have reached a reliable, quiet, end-to-end autonomous operation across the board. The distinction matters for security teams weighing immediate observed risks against predictions about how attacks may develop.
For organizations, the practical response is to prioritize known exploited vulnerabilities, restrict administrative access, reduce reliance on long-lived passwords and keys, and look for activity patterns unlikely to reflect ordinary human use. Those steps address established security risks as well as the automation described by Google; the report does not quantify how many incidents any specific measure would prevent.
Observed Activity Versus Google’s Forecast
Google’s report, titled GTIG AI Threat Tracker: Agentic Threat Activity Crosses a Significant Threshold, is a quarterly account of how criminal groups, state-backed actors and influence operations use large language models. Google said its findings drew on Mandiant incident-response work, threat-actor tracking and its platform defenses. The report’s central distinction is between activity researchers say they have observed and later stages they project.
In its seven-stage maturity model, Google labels the current state stage 5: minimal human involvement. It 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, three to six months. Those periods are Google’s forecast, not a finding that such operations have already become routine.
That difference helps explain conflicting headlines about the report. Coverage saying automated attacks have arrived refers to Google’s exploit_agent observation. Coverage saying fully automated attacks are coming soon reflects the projected later stages, in which a human might set a target and objective while agents execute a broader attack chain. Google has described increased stealth and sophistication as a still later stage.
“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
Scale and Operator Remain Unknown
Google did not say who operated the exposed server, how many websites were compromised, or when the activity took place beyond describing it as occurring in the past quarter. Its report says victim data suggested a focus on French media, entertainment and education websites, but does not provide a confirmed victim count or establish the full extent of the intrusions.
The report also does not show whether the WordPress plugin analysis led to an exploit. Nor does the available account establish that all the cited AI frameworks performed complete attack chains without human direction. Google’s stage 6 and 7 descriptions remain projections; the report does not provide a measured likelihood or a guarantee that its forecast timeline will hold.
Defenders Face Near-Term Decisions
Google’s report sets out a projected one-to-three-month window for attacks involving little to no human interaction, followed by a three-to-six-month window for fully autonomous attacks. These are estimates from the company’s maturity model, not announced launch dates or confirmed milestones. The report does not identify a specific future event that would verify when either stage has been reached.
For now, security teams can act on the observed risk by applying security updates, starting with actively exploited flaws; tightening administrative access; limiting long-lived credentials; and watching for unusually rapid or concentrated activity. Further clarity would depend on additional incident reports, evidence about the exploit_agent operator and victims, and whether researchers observe broader attacks that match Google’s projected stages.
Key Questions
What did Google say it found?
Google said it identified an exposed server running exploit_agent, a framework that compared software versions, inferred vulnerabilities fixed by updates, drafted exploits and automatically executed them.
Does the report say fully autonomous attacks are common now?
No. Google described the current stage as involving minimal human involvement. It reported one framework automating vulnerability discovery and exploitation, while broader attacks with little or no human interaction and fully autonomous attacks are projections.
Who was targeted by exploit_agent?
Google said data on the server suggested interest in French websites associated with media, entertainment and education. It did not identify the operator or give a confirmed count of affected websites.
What should organizations do in response?
Google’s report points to practical defensive steps: prioritize actively exploited software flaws, restrict administrative access, reduce long-lived credentials and monitor for unusually rapid bursts of activity.
Source: ThorstenMeyerAI.com