🔍 Read the full analysis: Google’s Warning On AI Attacks: What’s Evidence And What’s Forecast on ThorstenMeyerAI.com
Get smart everyday buys 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 framework that automatically identified and exploited software vulnerabilities without a person approving each step. The finding is one observed case; Google’s forecast that attacks with little or no human interaction will become more common in coming months is a projection, not a description of current widespread practice.
Google’s Threat Intelligence Group says it found an exposed server running an AI-assisted framework that automatically identified software vulnerabilities and launched exploits without a person approving each step. The November 16 report describes this as the first such activity Google has observed, but its forecast of attacks with little or no human involvement within months is a prediction, not evidence that fully autonomous attacks are already widespread.
The report, produced by Google Threat Intelligence Group (GTIG) with Mandiant, says the server hosted a multi-agent system called exploit_agent. Google says it could compare software code before and after an update, infer which security flaw the update fixed, draft an exploit and execute it. The system connected to multiple commercial and locally hosted large language models.
Google says the server also held stolen victim data and scripts for possible later activity, including network reconnaissance and planting web backdoors. The victim data suggested an interest in French websites in media, entertainment and education. Google did not identify the server’s operator or say how many sites were compromised. A report image shows the framework examining access checks in a WordPress plugin, but does not establish whether that examination resulted in an intrusion.
GTIG says it identified other AI-enabled tools in real-world intrusions, including the CAIRN framework against targets such as the U.S. aerospace sector and a customized PentAGI platform against organizations in Asia. These examples show use of AI tools in attacks, but they are distinct from the single exploit_agent finding that Google describes as fully automating vulnerability discovery and exploitation.
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 Attempts
The observed system matters because it links several steps that attackers have often carried out with human involvement: finding a likely vulnerability, building an exploit and running it. Automating that sequence could let an operator test more targets or act faster after software fixes become public. The report’s account of scripts for reconnaissance and backdoors also indicates that the server contained tools for activity beyond the initial exploit stage, though Google has not confirmed the extent to which those follow-on actions were carried out.
Google also warns that the current tools may sacrifice stealth for speed. The report describes them as fast but noisy, a tradeoff that can leave patterns defenders may detect. For organizations, the immediate implication is not that every attack is autonomous; it is that security teams should be prepared for rapid, machine-generated bursts of scanning or exploitation and should not wait for a human-paced sequence of activity.
Google’s recommendations include prioritizing actively exploited software flaws, reducing the use of long-lived passwords and keys, restricting administrative access and monitoring for unusually high-volume activity. These measures address exposure and detection risks whether an attacker uses AI or conventional tooling. The report does not quantify how many organizations face this specific framework or establish that any particular defense will prevent its use.
Products Worth Considering

Apple MacBook Air 15-inch Laptop with M5 Chip, 24GB Memory, 1TB SSD, Midnight

FLEXISPOT 3-in-1 Convertible Chair and Table for Four

Logitech MX Master 4 Ergonomic Wireless Mouse with Haptics – Graphite
As an affiliate, we earn on qualifying purchases.
Google’s Seven-Stage Threat Timeline
GTIG describes its report as a quarterly review of how criminal groups, state-backed actors and influence operations use large language models. It says its findings draw on Mandiant incident-response work, global threat tracking and platform defenses. That makes the report a mix of observed activity and the group’s assessment of how attacker capabilities may develop.
In its seven-stage timeline, Google labels the current point stage 5: “Minimal human-in-the-loop” and says it has been observed in the wild. Stages 6 and 7 are projected rather than established as routine. Google forecasts 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. In an October 30 post, GTIG described a later-stage model 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. The claim that automated attacks “have arrived” refers to Google’s observed exploit_agent case. The description “coming soon” reflects Google’s forecast of broader attacks that operate with little or no human interaction. The report supports the first as an observed instance, while presenting the second as a future possibility on Google’s timeline.
“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 Operators Remain Unknown
Google has not publicly identified who operated exploit_agent, how many websites were successfully compromised or whether the WordPress plugin analysis shown in the report led to an exploit. The report says victim data points to French media, entertainment and education sites, but does not provide a count or a complete account of affected organizations.
It is also unclear how often the framework was used, how reliably it worked, or whether the stolen data and follow-on scripts were connected to completed attacks. Google’s evidence establishes an observed system and its capabilities as described by the company; it does not show that fully automated exploitation is widespread. The forecasted stages remain uncertain, and the report does not give independent confirmation that attacks with little or no human input will become common on its stated schedule.
Defenders Face Near-Term Monitoring
Google’s report sets out no public deadline for a follow-up investigation or for identifying the server operator. Its near-term expectation is that attackers may use systems requiring less human input over the next one to three months, with more fully autonomous activity projected over three to six months. Those windows are GTIG’s forecasts, not confirmed milestones.
For now, organizations can act on the report’s defensive guidance: patch vulnerabilities known to be actively exploited, limit persistent credentials, secure administrator accounts and monitor for sudden bursts of scanning or exploitation. Further information about the server, affected sites and the frequency of similar activity would help establish whether the case marks a broader shift or remains an isolated example.
Key Questions
What did Google say it observed?
Google says it found an exposed server running exploit_agent, a multi-agent framework that could identify vulnerabilities by comparing software before and after a patch, draft exploits and execute them without a person approving each step.
Does Google say fully autonomous cyberattacks are already widespread?
No. Google describes one observed case of automated vulnerability discovery and exploitation. Its timeline labels the current stage as minimal human involvement; broader attacks with little or no human interaction are forecasts.
Who was targeted by exploit_agent?
Google says data on the server suggested interest in French websites in media, entertainment and education. It has not named the operator or reported how many sites were compromised.
When does Google expect more autonomous attacks?
GTIG forecasts attacks with little to no human interaction in the next one to three months and fully autonomous attacks in the three-to-six-month period. These are projections, not confirmed dates or guarantees.
What steps does Google recommend for defenders?
Google advises prioritizing patches for actively exploited flaws, reducing long-lived credentials, restricting administrative access and watching for unusually rapid bursts of activity. These are general security measures, not a guarantee against a particular attack.
Source: ThorstenMeyerAI.com
