Attackers Now Use AI Agents to Turn Patches Into Exploits, Google Says
Google's threat researchers found an attacker framework that reads security patches, writes exploits and runs them with no human involved. Why that makes monthly patching too slow for internet-facing systems, and what to do this week.

In this article
Table of Contents
Table of contents
Google says it has seen attackers use AI agents to turn security patches into working exploits and run them against victims with no human involved, the first time its threat researchers have recorded that. In a report published on Monday 16 November, the Google Threat Intelligence Group (GTIG) describes an exposed attacker server hosting a framework called exploit_agent that “compares code before and after patch announcements, analyzes findings to identify potential vulnerabilities, drafts exploits, and automatically executes them without human intervention.” There is no new vulnerability to patch here. What changes is how quickly a published fix becomes an attack.
#What Google found
GTIG says that in the past quarter it identified threat actors “using LLM frameworks to fully automate vulnerability discovery and exploitation without a human in the loop for the first time.” The exploit_agent server also held scripts for internal reconnaissance, brute-forcing passwords and installing webshells, and Google says data was taken from targeted victims. Its telemetry suggests the operator was interested in French websites in media, entertainment and education. Google does not name the operator; The Register describes a suspected Russian-speaking operator, based on the language of the tooling.
It is not a one-off. Google says it has seen several groups experimenting with similar frameworks, including financially motivated criminals and suspected espionage actors linked to China.
#How short the gap has become
Google’s clearest example is PaperCut, the print management software. In late August two PaperCut NG/MF flaws (CVE-2026-81578 and CVE-2026-82078) were exploited as zero-days. Security researchers, including Huntress, built working exploits within a day of disclosure, and Google cites GreyNoise as seeing a campaign four days after disclosure in which an attacker likely used AI models to build and test exploits and hit hundreds of organisations, going from first foothold to domain administrator within minutes. Google also counts zero-day exploits rising from an average of 8 a month in 2025 to 22 in August 2026.
#Monthly patching is now too slow for anything on the internet
In our view this is the practical meaning of Google’s report. Reading a patch to work out what it fixed has always been how attackers built exploits for known flaws; it took skill and days or weeks. If an agent does it unattended, the time between a vendor’s fix and attacks on unpatched systems shrinks to hours or a few days, and it costs the attacker almost nothing to try every patch that comes out. A monthly patch cycle still suits most internal systems. It no longer suits VPN gateways, remote access tools, print servers, file transfer and web applications that face the internet.
We think there is a second, smaller effect. Google says that after PHP’s October fixes “hundreds of claimed but non-functional POCs, likely generated with the help of LLMs, were published.” Expect more fake exploit code on GitHub, some of it carrying malware, aimed at the admins who download it to test their own systems.
#What to do this week
- List every system you expose to the internet and give critical fixes for those a target of days, not the next monthly window. Where you cannot patch that fast, put the service behind a VPN or zero-trust access, or switch it off until you can.
- If you run PaperCut NG or MF, confirm PaperCut’s fixes for CVE-2026-81578 and CVE-2026-82078 are installed and that the admin interface is not reachable from the internet.
- Do not run proof-of-concept code from public repositories against your own systems unless your security team has read it first.
Evidence trail
Sources and verification
Primary documentation and external technical references used in this article.
- 01Google Threat Intelligence Group: Agentic threat activity crosses a significant thresholdcloud.google.com
- 02The Register: Google says AI agent turned a security patch into a working exploit with no human helptheregister.com
- 03The Hacker News: Google finds hackers using AI agents to build and run exploits from software patchesthehackernews.com
About this page
Credited to
Dates
Published
Corrections
No corrections are listed for this page. Some review sweeps corrected pages without naming each one; those are in the corrections log.
Read-only scripts
- Windows devices with no BitLocker recovery key in Entra IDMicrosoft Entra ID, Windows
- Intune devices that haven't synced in N daysMicrosoft Intune
- Mailboxes forwarding to external addressesExchange Online
- BitLocker status of every volume on this PCWindows
Upcoming deadlines
- Intune moves to macOS 15 and laterQ4 2026
- SAP SuccessFactors provisioning: move from basic auth to workload identityNovember 2026
- Intune November release: JIT registration reminder at APNs renewal and automatic JIT for eligible tenantsNovember 2026
Explore KBY
The KBY Brief
A short email about twice a week with new runbooks, configuration traps and fixes from KBY Technologies.
SubscribeMore news
All news
Platforms and devicesMicrosoft Previews Remote Lock for Lost Windows Devices in Intune


Was this useful?
Reports go to the site owner and are used to correct pages. Corrections are listed at /corrections.
Get new fixes by email.
New runbooks, configuration traps and fixes from KBY Technologies, about twice a week.
Comments
Add a thoughtful note on Attackers Now Use AI Agents to Turn Patches Into Exploits, Google Says. Comments are checked for spam and held for moderation before appearing.