Skip to main content
Live
Main content

Google says exploit_agent AI framework turned patches into exploits without humans

The framework, found on an exposed staging server, compares code before and after patches and fires exploits automatically, per Google's latest AI Threat Tracker.

Jaeden Schafer
Editor in Chief · · 5 min read
Google says exploit_agent AI framework turned patches into exploits without humans

Google Threat Intelligence Group says it has seen threat actors use large language model frameworks to fully automate vulnerability discovery and exploitation, with no human in the loop, for the first time. The clearest example is a tool GTIG calls exploit_agent, a multi-agent vulnerability research and weaponization framework found on an exposed staging server and wired into multiple frontier and locally hosted models, according to the group's AI Threat Tracker report published November 16.

Google says exploit_agent runs continuous differential analysis of popular web application software. It compares code before and after patch announcements, analyzes the changes for likely vulnerabilities, drafts exploits and executes them automatically. The same server held data exfiltrated from victims, plus scripts for internal reconnaissance, brute-forcing, installing webshells and removing other attackers' webshells.

Victim telemetry shows the operator was specifically interested in French websites tied to media and entertainment, as well as the education sector. Comments in a configuration file were written in Russian. In GTIG's translation, one labels the entries as LLM API keys for an "exploit_agent pilot" and tells the user to enter the keys "right here, not in the chat." Google did not name the operator or tie exploit_agent to a known threat group.

Key facts

  • 01Google's GTIG found exploit_agent, an AI framework that turns patch diffs into executed exploits without human intervention.
  • 02The framework connects to multiple frontier and local LLMs and targeted French media, entertainment and education websites.
  • 03Zero-day exploits rose from 8 per month in 2025 to 11 per month in 2026, spiking to 22 in August, per GTIG.
  • 04GreyNoise saw a PaperCut exploitation campaign hit hundreds of organizations four days after disclosure.
  • 05A contractor used the Gemini API to build a platform to manage 100,000+ accounts; Google disabled the assets.

The finding marks a shift from GTIG's previous quarterly edition, published roughly two months earlier, when the group said it had not observed fully autonomous pipelines in the wild. Google now assesses that attackers may find it more efficient to weaponize n-day vulnerabilities, flaws that already have public patches, than to hunt for zero-days.

Google's numbers show why patch timing matters. Overall vulnerability disclosures and exploits nearly doubled between January and August 2026, GTIG wrote, while zero-day exploits rose from a 2025 average of 8 per month to 11 per month in 2026 before spiking to 22 in August.

PaperCut is the report's case study. Two chained flaws in PaperCut print management servers, CVE-2026-81578, an authentication bypass, and CVE-2026-82078, unsafe dynamic class loading, enabled unauthenticated remote code execution in late August. Huntress and other researchers had working exploits within a day of disclosure. GreyNoise reported a campaign four days after disclosure that hit hundreds of organizations, and in one case an attacker expanded a foothold to domain administrator within minutes.

GTIG also flagged two PHP vulnerabilities, CVE-2026-109618 and CVE-2026-109619, which were discovered with LLM-enabled analysis and patched on October 21, 2026. A proof-of-concept video and early patched code appeared before the official fix, followed by hundreds of claimed proof-of-concept exploits that did not work.

Beyond exploit_agent, Google tracked attackers adopting open-source agentic security frameworks, including ARTEX, CAIRN, CyberStrikeAI and PentAGI, along with experimentation with HexStrike AI. GTIG observed CAIRN in intrusion activity against targets including the US aerospace sector, and tracked a customized attack platform built on the open-source version of PentAGI scanning organizations in Asia from late August to early October.

Related · from this week
Google says it disrupted the first AI-built zero-day exploit in the wild
Jaeden Schafer · 4 min read →

The report also covers influence operations built on Google's own models. A commercial contractor used the Gemini API to build a platform designed to manage more than 100,000 accounts across major global platforms. Google disabled the accounts and assets tied to that misuse, and Google DeepMind used the findings to strengthen its classifiers.

On the defensive side, Google said its SecOps team deployed rulepacks to detect ARTEX, and Mandiant Consulting released a hardening guide and an architectural defense framework. Google also cited its own AI-driven security work, including Big Sleep and CodeMender, alongside its Vulnerability Reward Program and Project Zero.

ShareXLinkedInEmail
AI Box

Every AI model. One chat.

The latest models from ChatGPT, Claude, Gemini, Sora, ElevenLabs — 80+ models in a single chat. Compare answers side by side. Pick the best one every time.

  • ChatGPT, Claude, Gemini, Grok, DeepSeek — in one chat
  • Generate images & video with Sora, Veo, Ideogram
  • Compare any two models side by side
  • From $8.99/mo · 80+ models, all included
Try AI Boxaibox.ai
Trusted by 3,000+ teams
Got a tip?

Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.

Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.

AI Box Daily briefingFree · Daily · No fluff

Stay ahead of everyone in AI.

The tightly edited AI news email engineers, founders, and investors actually open. One email. Every weekday. Five minutes to finish.

Loved by 10,000+ AI professionals
Free forever. Unsubscribe with one click.

The briefing read inside teams at

Keep reading

More from Security

Google logo
Security

Google says it disrupted the first AI-built zero-day exploit in the wild

GTIG flagged a 'hallucinated CVSS score' and textbook formatting in a Python exploit targeting 2FA on an open-source admin tool.

Jaeden Schafer4 min read
Google logo
Security

Google's Gemini autonomously hacked three companies in Irregular security test

The model guessed passwords in one breach and pulled credentials from public repositories in two others before halting on its own.

Jaeden Schafer4 min read
Google logo
Security

Google fixes 1,072 Chrome bugs in one month using AI, more than the past two years combined

Chrome 149 and 150 patched more security flaws than the previous 23 releases combined, as Google credits Gemini for the surge.

Jaeden Schafer5 min read