AI Attacks Without Humans In The Loop: Reading Google’s Findings Carefully
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: AI Attacks Without Humans In The Loop: Reading Google’s Findings Carefully on ThorstenMeyerAI.com

Before you orderOffer from Amazon

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
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

TL;DR

Google’s Threat Intelligence Group says it found an exposed server running a framework that automated vulnerability research and exploitation without a person approving each step. The finding concerns at least one observed case; Google’s prediction that attacks with little or no human involvement will become more common in coming months is a forecast, not a description of current widespread activity.

Google’s Threat Intelligence Group says it found an exposed server running software that used AI agents to identify vulnerabilities in updated code, write exploits and execute them without a person approving each step. The November 16 report describes this as the first time Google has seen attackers fully automate vulnerability discovery and exploitation, but it does not establish that such attacks are widespread or that the system succeeded against every target it examined.

Google said the server hosted a multi-agent framework it called exploit_agent, connected to several commercial and locally run AI models. According to the report, it compared software code before and after security patches, inferred which weaknesses the updates addressed, then drafted and ran attacks. Google also found stolen victim data and scripts for possible later activity, including network exploration and installing web backdoors.

The available evidence leaves the scale and ownership unclear. Google did not identify who operated the server or say how many websites were compromised. Data on the server, the company said, suggested an interest in French websites in media, entertainment and education. The report also illustrates the framework examining access checks in a WordPress plugin, but does not say that this examination led to a successful attack.

Google’s report describes other use of AI-enabled tools in intrusions, including frameworks it names as CAIRN and PentAGI. It says it observed CAIRN used against targets including the U.S. aerospace sector and a customized PentAGI platform used against organizations in Asia. These examples are separate from the exploit_agent server finding and do not, on their own, establish that those operations were fully autonomous.

At a glance
reportWhen: Report published November 16, 2026; fin…
The developmentGoogle published a threat report describing an observed AI framework that automated vulnerability discovery and exploitation, while predicting that more autonomous attacks could follow.

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

  1. 1Prompting a chatbothuman-driven
  2. 2AI built into malwareautomated runtime call
  3. 3AI automation in operationsAI in the loop, humans in charge
  4. 4Experimenting with autonomous frameworksexperimental autonomy

Now · in the wild

  1. 5AI tools run several attack stages“minimal human-in-the-loop”

Projected · forecast, not observed

  1. 6Simple attacks with little to no human interactionnear-term, 1–3 months
  2. 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.

Google’s own label for today: stage 5 of 7. Fully automated exploitation exists in at least one case. Fully autonomous attacks as the norm are still a forecast.

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 Has Reached Exploit Work

The report matters because the observed system did more than use an AI model to suggest code or summarize a vulnerability. Google says it linked vulnerability analysis, exploit generation and execution in an automated process. If such tools are adopted more broadly, attackers could examine patches and attempt exploitation faster, potentially reducing the time defenders have to respond.

That potential is not the same as proof of a broad shift in attacks. Google says the tools currently trade stealth for speed: their activity can be noisy and easier to detect. The report does not quantify how often exploit_agent was used, how many attacks it completed, or whether its results were more effective than work directed by people. For security teams, the immediate relevance is to treat rapid exploitation and unusual activity bursts as risks, without treating a forecast as an established norm.

Products Worth Considering

Bose TV Speaker (2nd Gen) Compact Soundbar with Bluetooth and HDMI Connectivity

Bose TV Speaker (2nd Gen) Compact Soundbar with Bluetooth and HDMI Connectivity

As an affiliate, we earn on qualifying purchases.

Observed Activity Versus Forecast

Google published the report, titled GTIG AI Threat Tracker: Agentic Threat Activity Crosses a Significant Threshold, with Mandiant on November 16, 2026. It says the analysis draws on incident-response work, threat-actor tracking and live platform defenses. Google’s headline finding is that it identified attackers using large-language-model frameworks to automate vulnerability discovery and exploitation without a human in the loop for the first time.

The report distinguishes that observation from a staged timeline of increasingly autonomous attacks. Google places current activity at stage 5 of seven, described as involving 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 stages are forecasts, not confirmed events.

This distinction helps explain differing coverage of the report. Headlines saying automated attacks have “arrived” refer to the exploit_agent case; headlines saying they are “coming soon” refer to Google’s expected progression toward broader autonomy. The evidence supports the narrower point that Google observed one framework automating key exploit steps. It does not show that fully autonomous attacks are already routine.

“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 Success Remain Unknown

Google has not disclosed the operator’s identity, the number of victims, or how many attacks the framework completed. Its description of the server’s data indicates a possible targeting interest in French sites, but does not confirm the number of affected organizations or the outcome of each attempted exploit. The WordPress example likewise shows an inspection of access checks, not a confirmed compromise.

The report also does not provide enough detail to establish how much human involvement occurred outside the specific automated steps. Google labels the broader current stage as involving minimal human involvement, while reserving fully autonomous attacks for a projected future stage. Its timeline is an expectation; the report does not establish that the predicted schedule will be met.

Defenders Face Immediate Patch Work

Google’s report points security teams toward established defensive measures: prioritize actively exploited vulnerabilities, shorten the life of credentials and keys, restrict administrative access, and monitor for bursts of activity unlikely to reflect ordinary human behavior. These steps address the risks described without assuming that every organization is already facing fully autonomous attacks.

The next evidence to watch for is whether Google or other incident responders report additional cases, disclose confirmed victim counts, or document successful attacks linked to similar frameworks. Google’s one-to-three-month and three-to-six-month windows are forecasts, so readers should look for reported activity and updated evidence rather than treat the timeline as a confirmed schedule.

Key Questions

What did Google say it found?

Google said it found an exposed server hosting exploit_agent, a framework that analyzed code changes after security updates, inferred vulnerabilities, drafted exploits and automatically executed them. The report does not establish how many attacks succeeded.

Does the report show that fully autonomous attacks are widespread?

No. Google describes an observed framework automating vulnerability discovery and exploitation, but says current activity remains at a stage with minimal human involvement. Broader, fully autonomous attacks are a forecast, not a confirmed widespread practice.

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 provide a count of confirmed victims.

What does Google expect next?

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 three to six months. These are estimates, not guarantees.

What should organizations do now?

Google’s recommendations include prioritizing actively exploited flaws, limiting long-lived passwords and keys, tightening administrative access, and monitoring for unusually rapid bursts of activity.

Source: ThorstenMeyerAI.com

You May Also Like

How To Understand The Claude Code VS Code Extension SSH Backdoor Incident

A report alleges the Claude Code VS Code extension installed a persistent SSH backdoor, but the supplied material lacks technical evidence and vendor confirmation.

Claude Duet’s Move Into Always-On AI Puts Coordination On The Agenda

Anthropic launched Claude Duet as a waitlisted research preview, with a cloud computer for ongoing tasks and approval requirements for sensitive actions.

Epoch AI’s Firefox Test: Understanding The Attacks And What To Do Now

Epoch AI says GLM 5.3 built attacks against six Firefox releases in a lab test. Mozilla patched a related flaw in Firefox 159.

What’s Driving China’s Objection To SoftBank’s Factory Robot Deal?

ABB says China prohibited SoftBank’s purchase of its robotics business in China. The global deal’s status and Beijing’s reasons remain unclear.