Incident sheet · /security
- System
- Security & Risk
- Filed
- Read
- 3 min · 522 words
- Severity
- Routine
Google: Fully Autonomous AI Cyberattacks Expected Within Six Months
Google GTIG projects fully autonomous agentic attacks within six months and little-to-no-human-interaction attacks within three, citing exploit_agent and ARTEX abuse.

Event log
- Google projects fully autonomous agentic attacks within six months; little-to-no-human-interaction attacks within three months.
- GTIG observed the first fully automated vulnerability discovery and exploitation without a human in the loop in the past quarter.
- The FBI warned ARTEX has been weaponized since August against financial, government, technology, defense and healthcare targets.
- Hundreds of mostly AI-generated PoC exploits appeared shortly after PHP announced two flaws in the PHP2Shell case.
- UNC6836 used modified Claude Code to deliver the SOMBERMEME backdoor to cryptocurrency firms.
Google Threat Intelligence Group projects fully autonomous AI-driven attacks within six months. In the nearer term, the company expects unsophisticated attacks requiring little to no human interaction within three months, relying mostly on multiple AI agents (autonomous software systems that plan and execute tasks).
GTIG reported the assessment in its AI Threat Tracker. The group has observed large-scale adoption of AI-based offensive workflows by both financially motivated and state-sponsored threat groups. AI moved from experimental use to routine deployment.
What has already been deployed?
Attackers now systematically abuse open-source penetration-testing suites to run fully automated attacks. Google names four frameworks in active abuse: ARTEX, CAIRN, CyberStrikeAI, and HexStrike AI.
One documented case involved CAIRN, an open-source multi-agent penetration-testing framework, used in an attack against the U.S. aerospace sector. Suspected North Korean hackers used a modified Claude Code to deliver malware.
GTIG highlighted a multi-agent system dubbed 'exploit_agent' that fully automated vulnerability discovery and exploitation against French media, entertainment, and education websites.
"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 said.
How does exploit_agent work?
Google discovered the framework on an exposed staging server belonging to the attacker. The server also held stolen victim data, reconnaissance information, webshells, and brute-forcing scripts.
"More specifically, GTIG identified an exposed staging server hosting a multi-agent vulnerability research and weaponization framework, dubbed exploit_agent, that integrates with multiple frontier and local LLMs," Google said.
The system analyzes software updates on popular web applications to find signs of security flaws. It then automatically attempts to exploit them. Attackers used such frameworks for adaptive attacks against internet-exposed endpoints vulnerable to low-complexity flaws for which large language models could easily build exploits.
What does this mean for defenders?
Google warns these frameworks have dramatically decreased the time defenders have to patch systems. They also generate false leads that burden security teams.
The PHP2Shell case illustrates the problem. Hundreds of purported proof-of-concept exploits, mostly AI-generated, appeared shortly after PHP announced two flaws. Defenders had to verify each piece of code and develop protections.
What did the FBI and other reporting find?
The FBI warned in an advisory last week that multiple threat actors have weaponized the ARTEX AI pentesting framework since August. Targets spanned the financial, government, technology, defense, and healthcare sectors.
Google separately tracked UNC6836, an activity cluster with suspected links to Midnight Neptune, attacking cryptocurrency firms in North America and Europe. After initial compromise, the attackers modified the Claude Code app's settings so that launch triggered download of the SOMBERMEME backdoor.
The researchers also report AI is now routinely used in influence operations by state-backed groups, commercial vendors, and political networks. LLMjacking, the theft of language-model access for resale or abuse, has become a widespread problem. Attackers increasingly exploit this revenue stream at significant cost to affected organizations.
Enterprises should prioritize patching internet-exposed endpoints with low-complexity vulnerabilities, the class of flaws these frameworks target most effectively.
via bleepingcomputer.com (Source)
More from Rebecca Stone
Show full bio
Senior reporter covering money, funding and market shifts around enterprise information technology and the leaders who run at CIO Monitor.
199 articles
The newsroom digest
One email a day with the stories that matter. No noise.
Cross-references
- Google Finds AI Agents Turning Patches into Working Exploits Autonomously
- FBI Warns AI Hacking Framework ARTEX Drives Global Attack Wave
- Microsoft's CISO Wants AI Pen-Testing Framework Regulated. Security Leaders Say It's Too Late
- FBI: Weaponized ARTEX Used in Healthcare and Other Sector Attacks
- Wiz AI Scan Finds More Than 475 High-Severity and Critical Exposures