🔍 Read the full analysis: AI Attacks Without A Human In The Loop: A Closer Look At Google’s Claims on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
Google’s Threat Intelligence Group says it found an exposed server running a framework that automatically examined software updates, developed exploits and executed them without human approval at each step. Google describes this as a first observed case, but its forecast of attacks with little or no human interaction in coming months is a projection, not a claim that such attacks are already widespread.
Google’s Threat Intelligence Group says it found an exposed server running software that could identify vulnerabilities from security updates, write exploits and run them automatically, without a person approving each step. The finding, published November 16 with Mandiant, is a reported example of automated exploitation—not evidence that fully autonomous cyberattacks are already widespread. Google forecasts that attacks involving little or no human interaction could follow in the coming months.
In its report, “GTIG AI Threat Tracker: Agentic Threat Activity Crosses a Significant Threshold,” Google says the server hosted a multi-agent framework it named exploit_agent, connected to commercial and locally run large language models. The system compared software code before and after updates, inferred what weakness a patch addressed, drafted an attack and automatically executed it, according to the company. Google describes this as the first time it had observed threat actors using LLM frameworks to fully automate vulnerability discovery and exploitation without a human in the loop.
Google says the server also contained stolen victim data and scripts for follow-on activity, including network reconnaissance and installing web backdoors. Data on the server suggested an interest in French websites in media, entertainment and education. The report does not identify the operator or say how many sites were compromised. It also does not establish that a pictured examination of a WordPress plugin’s access checks led to a successful attack.
The report describes other AI-enabled tools in use, including open-source penetration-testing frameworks. Google says it observed CAIRN used against targets that included the U.S. aerospace sector, and a customized PentAGI platform used against organizations in Asia. These examples concern frameworks used in real intrusions; they are separate from the specific exploit_agent server finding.
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 Automated Exploits Matter
The reported system could shorten the time between a software vendor releasing a fix and an attacker attempting to exploit the flaw it addresses. By comparing versions, a tool can use the patch itself as a clue to what changed. That makes prompt patching of actively exploited vulnerabilities especially relevant for organizations that rely on exposed web software.
Google also cautions that speed does not necessarily mean stealth. The report says current tools “trade stealth for velocity”: automated activity may be fast and noisy, potentially giving defenders signals to detect. This is a measured assessment, not a guarantee that every automated attack will be easy to spot. The practical concern is that security teams may face more attempts in less time, while the techniques and visibility vary by case.
For companies and public institutions, the finding supports familiar defensive priorities rather than proving that existing safeguards no longer work: apply security updates, restrict administrative access, reduce reliance on long-lived credentials and monitor for unusual bursts of activity. Google’s account makes the issue concrete, but does not quantify the scale of resulting compromises or the rate of successful attacks.
Popular right now

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

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation

Roborock Q10 S5+ Robot Vacuum and Mop with Self-Emptying Station
As an affiliate, we earn on qualifying purchases.
Observed Activity Versus Forecast
The difference between headlines saying automated attacks have “arrived” and those saying they are “coming soon” reflects two distinct parts of Google’s report. The exploit_agent case is described as observed activity: Google says one system automated vulnerability discovery and exploitation. Its broader maturity timeline, by contrast, places the current environment at stage five of seven, labeled “minimal human-in-the-loop,” and marks later stages as projected.
Google projects attacks with little to no human interaction in the near term, which it defines as one to three months, and fully autonomous attacks in the medium term, three to six months. In an earlier October 30 post, the group described a later stage in which a human sets the target and objective while multi-agent systems carry out the attack chain. These are Google’s forecasts and maturity model, not confirmed milestones that every attacker will reach on schedule.
The report draws on Google’s threat tracking, platform defenses and Mandiant incident-response work, according to the company. That gives the findings a basis in its own security operations, but the public account does not provide enough information to independently assess the exploit_agent server, its operator or the number of affected organizations.
““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 named the server’s operator, said how many websites were breached, or detailed the confirmed effects on victims. The report’s description of data that “suggests” interest in French media, entertainment and education sites does not establish that all, or any particular number, of those organizations were successfully attacked. The available account also does not show whether the WordPress plugin analysis resulted in exploitation.
It remains unclear how broadly exploit_agent or comparable systems are being used, how often their attempts succeed, and whether the activity will develop along Google’s projected timeline. Google’s finding supports the claim that it observed at least one automated exploitation system; it does not establish that attacks with no meaningful human involvement are already routine.
Defenses and the Next Evidence
Google’s report advises defenders to prioritize actively exploited software flaws, limit long-lived keys and passwords, tighten administrative access and look for bursts of activity that could indicate automation. These steps address risks raised by the report, although Google has not supplied a public count of victims or a measure of how much each step reduces exposure.
The next evidence to watch for is further reporting that clarifies the server’s operator, affected organizations and whether similar systems appear in additional investigations. Google’s one-to-three-month and three-to-six-month projections are time windows in its model, not announced deadlines. Until more evidence is available, the clearest distinction remains: one automated exploitation case was reported as observed; broader fully autonomous attacks remain a forecast.
Key Questions
What did Google say it found?
Google says it identified an exposed server hosting exploit_agent, a multi-agent system that compared software versions, inferred vulnerabilities from patches, drafted exploits and ran them automatically.
Does the report say fully autonomous cyberattacks are widespread?
No. Google reports an observed case of automated vulnerability discovery and exploitation. Its timeline places the current stage at minimal human involvement and treats later, more autonomous stages as projections.
Were particular French organizations confirmed as victims?
The report says data on the server suggested interest in French media, entertainment and education websites. Google did not state how many sites were compromised or confirm specific victims in the source material.
What should organizations do in response?
Google recommends prioritizing actively exploited flaws, limiting long-lived credentials, restricting administrative access and monitoring for unusual bursts of activity. The report does not quantify the effect of these measures on risk.
Source: ThorstenMeyerAI.com
