🔍 Read the full analysis: When AI Attacks Have No Human In The Loop: Google’s Findings In Context 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 November 16 threat report describes an exposed server running exploit_agent, a framework that it says automatically identified and exploited software vulnerabilities without a person approving each step. Google calls this a first observed case, but says broader attacks with little human involvement are projected for the coming months; the scale and operator behind the server remain unknown.
Google Threat Intelligence Group says it found an exposed server running software that automatically analyzed security updates, developed exploits for the vulnerabilities they fixed and executed those exploits without human approval at each step. The November 16 report describes this as the first case Google has observed of attackers fully automating vulnerability discovery and exploitation, while warning that more autonomous attacks remain a forecast rather than an established norm.
The report, published with Mandiant under the title “GTIG AI Threat Tracker: Agentic Threat Activity Crosses a Significant Threshold,” says the server hosted a multi-agent framework Google called exploit_agent. The framework connected to multiple commercial and locally run large language models. According to Google, it monitored widely used web software, compared code before and after updates, inferred which weaknesses a patch addressed, drafted attacks and ran them automatically.
Google says the server also contained data taken from victims and scripts for further operations, including network reconnaissance and placing web backdoors. Victim data suggested the operator was interested in French websites in media, entertainment and education. The report does not establish how many sites were compromised or identify the operator. It also shows the framework examining access checks in a WordPress plugin, but does not say whether that examination resulted in an attack.
The report describes other uses of AI-enabled tools in intrusions. Google says it observed the CAIRN framework used against targets including the U.S. aerospace sector, and a customized PentAGI platform used against organizations in Asia. Those findings indicate that AI-assisted offensive frameworks are not limited to laboratory tests, although they do not mean every tool performs an entire attack autonomously.
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
A Faster Exploit Cycle, With Limits
Automating the path from a software update to an exploit could shorten the time defenders have to patch vulnerable systems. A system that can infer what a fix changes and quickly test an attack may let operators probe more software and launch attacks faster than a person working through each step. The risk is especially relevant for organizations running exposed web applications and plugins.
Google also describes a constraint: current systems can trade stealth for velocity. Rapid, repetitive activity may stand out in security monitoring, making automated campaigns easier to detect than slower operations. The report does not quantify how often exploit_agent succeeded, how many systems it reached or whether its approach led to lasting access. Its finding points to a capability and a potential shift in attacker workflows, not proof that autonomous attacks are already widespread.
Products Worth Considering

FLEXISPOT EN1 Electric Standing Desk, 48 x 24 in, Black

Canon SELPHY CP1500 Compact Photo Printer Bundle with Two KP-108IN Ink and Paper Sets, Black

Canon SELPHY CP1500 Wireless Compact Photo Printer with RP-108 Ink and Paper Set, Black, 108 Sheets
As an affiliate, we earn on qualifying purchases.
Observed Automation Versus Forecasts
Google’s report is a quarterly assessment of criminal, state-backed and influence operations involving large language models. It says its findings draw on Mandiant incident-response telemetry, threat-actor tracking and Google’s live platform defenses. Its central distinction is between activity observed in the field and later stages that it expects attackers to reach.
Google places current activity at stage five of a seven-stage timeline, labeled “minimal human-in-the-loop.” It projects attacks with little or 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, Google described a later stage in which a person sets a target and objective while multi-agent systems carry out the attack chain. These are Google’s projections, not findings that those later stages are already typical.
That distinction explains differing coverage of the report. The observed exploit_agent case supports saying that fully automated vulnerability discovery and exploitation has appeared in at least one case identified by Google. The projected timeline supports saying that broader, more autonomous attack chains may become more common in coming months. Neither claim establishes that fully autonomous attacks are now the standard method used by attackers.
“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
The Server’s Reach Remains Unknown
Google has not disclosed who operated exploit_agent, how many websites were breached, or the full extent of any damage. The report’s reference to victim data indicates information was present on the server, but does not provide a complete count of affected organizations or explain how the data was obtained. The WordPress plugin example likewise does not establish that the tested access check was exploited.
It is also unclear how reliable the framework’s exploit generation was, whether a human selected targets or configured the system, and how often the automated process succeeded. Google’s timeline for attacks with little or no human interaction is a forecast, and the report does not provide a measured rate at which attackers are moving through its stages. The single server finding should not be read as evidence that autonomous attacks are widespread.
Defenders Face Near-Term Patching
Google’s report does not announce a specific enforcement action or give a date for a follow-up investigation. For defenders, the immediate response is to prioritize actively exploited vulnerabilities, restrict administrative access and reduce long-lived passwords and keys, as the source report recommends. Security teams can also look for sudden bursts of scanning or exploitation activity that appear faster or more repetitive than a person would typically produce.
Further reporting or incident disclosures may clarify the server’s operator, targets and consequences. Google’s next one-to-three-month projection is for attacks with little to no human interaction; its three-to-six-month projection is for fully autonomous attacks. Those windows are estimates, not confirmed deadlines. Whether the predicted capabilities become routine—and whether they can operate with greater stealth—remains to be seen.
Key Questions
What did Google say it found?
Google says it identified an exposed server hosting exploit_agent, a multi-agent framework that analyzed software patches, developed exploits for the vulnerabilities they fixed and executed them automatically.
Does the report prove fully autonomous cyberattacks are widespread?
No. Google describes one observed server and framework that automated vulnerability discovery and exploitation. Its broader forecast for attacks with little or no human interaction is presented as a future development, not as a widespread current practice.
Who was targeted by exploit_agent?
Google says data on the server suggested interest in French media, entertainment and education websites. It has not publicly identified the operator or provided a count of compromised sites.
What does Google expect next?
Google projects attacks with little to no human interaction in the next one to three months, followed by fully autonomous attacks in three to six months. These are forecasts, not confirmed milestones.
What can organizations do now?
The source report recommends prioritizing actively exploited software flaws, limiting administrative access, reducing long-lived credentials and monitoring for unusually rapid bursts of activity.
Source: ThorstenMeyerAI.com
