@readmodelmag
MODEL

Tools

Google caught a fully automated AI attack. It didn't need a zero-day. It read the patch.

SummaryGoogle's latest AI Threat Tracker describes exploit_agent, a multi-agent system that diffs web software before and after each patch, drafts an exploit and runs it with no human in the loop. The same report counts hundreds of AI-written exploits that don't work. Our take: sloppy AI code now hits defenders from both sides.

· MODEL Editors · 2 min read

Some links may be affiliate links. Disclosure

MODEL card on a cream background. Small red kicker: HOT TAKE / AI SECURITY. Black headline: AI now reads your patch, then attacks. Bottom left: MODEL. Bottom right: Sources: Google Threat Intelligence Group, Nov. 16, 2026.
MODEL card for this story. Illustration by MODEL. Original. Image source.

What happened. On Nov. 16, 2026, Google Threat Intelligence Group (GTIG) and Mandiant published their quarterly AI Threat Tracker. Its headline finding: 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 Cloud Blog (opens in a new tab); CyberScoop (opens in a new tab); BleepingComputer (opens in a new tab); Computer Weekly (opens in a new tab)).

exploit_agent, in Google's words. GTIG found it on an exposed staging server:

  • A multi-agent framework that works with multiple frontier and local LLMs.
  • It runs continuous differential analysis of popular web application software: it compares code before and after patch announcements, finds the likely flaw, drafts an exploit and executes it automatically.
  • Alongside it: scripts for internal reconnaissance, brute-forcing, installing webshells and removing other attackers' webshells.
  • Victim telemetry pointed to French websites in media, entertainment and education.
  • Comments in its environment file were in Russian. Translated by Google: "Enter the keys yourself, right here, not in the chat."

The sloppy-code side. Google also says the lower barrier "has fueled a wave of LLM-generated exploits that are claimed to work but don't." After official patches for two PHP flaws (CVE-2026-109618 and CVE-2026-109619) landed on Oct. 21, 2026, hundreds of claimed but non-functional proofs of concept, likely written with LLMs, were published within days, and the security community had to verify or debunk them.

Agents as the way in. In intrusions at cryptocurrency firms in North America and Europe, a suspected North Korean cluster (MIDNIGHT NEPTUNE) pivoted into internal code repositories and modified .claude/settings.json files, so that starting Claude Code triggered a download of the SOMBERMEME backdoor.

The numbers. Google says overall disclosures and exploits nearly doubled between January and August 2026. Exploited zero-days went from an average of 8 a month in 2025 to 11 a month in 2026, then 22 in August.

Our take. AI-written code is now sloppy and fast on both sides. Defenders spend hours debunking fake exploits, while the real ones are built from the patch diff itself. The time between "fix released" and "fix exploited" is now however long it takes an agent to read your changelog.

What to do:

  1. Patch internet-facing web apps and plugins the day the fix ships. Google says attackers find already-patched (n-day) flaws the most efficient target.
  2. Treat agent config files (.claude/settings.json and similar) as code: require review on every change and alert on edits.
  3. Before you trust a public proof of concept, assume it may be AI-generated and broken. Test it in isolation.

MODEL Editors. Not sponsored. MODEL did not test these attacks.

Get three tested tools every Friday

Free, weekly. Unsubscribe anytime.

Sources

  1. GTIG AI Threat Tracker: Agentic Threat Activity Crosses a Significant Threshold

    Google Cloud Blog (Google Threat Intelligence Group, Nov. 16, 2026)

  2. The era of autonomous attacks has arrived.

    The Keyword (Google)

  3. Google says Russian-speaking hackers ran the first fully automated AI cyberattack

    CyberScoop

  4. Google warns that fully-automated AI attacks are coming soon

    BleepingComputer

  5. Threat actor use of AI becoming routine, warns Google

    Computer Weekly

FREE · EVERY WEEK

The top AI tools this week, tested.

Every week MODEL's editors pick the AI tools worth your time, try them hands-on, and tell you what's good, what's not, and who it's for. Free, in your inbox.

Free. Unsubscribe anytime. Our picks are editorial and can't be bought.