Google Catches Hackers Using AI to Turn Patches Into Exploits

According to a new threat intelligence report published by Google, hackers are no longer just using artificial intelligence to write phishing emails, but are deploying fully autonomous multi-agent AI frameworks to hack into target networks at “machine speed.”

Infographic of an orchestration agent with five stages: reconnaissance, initial access, persistence, internal movement, and complete objectives.

The report highlights how cybercriminals and state-sponsored espionage groups are replacing single-prompt interactions with interconnected AI pipelines. These agentic frameworks continuously scan target networks, conduct patch differential analysis, craft custom exploits, and execute multi-stage intrusions.

While security researchers previously observed hackers experimenting with basic AI tools, Google Threat Intelligence Group’s (GTIG) telemetry reveals that threat actors are now actively operationalizing open-source, multi-agent penetration testing frameworks during live intrusions.

These autonomous systems leverage specialized AI sub-agents that collaborate via shared decision-making loops to navigate target networks:

  • CAIRN: An open-source framework utilizing an “Observe, Orient, Decide, Act” (OODA) loop. GTIG confirmed CAIRN was recently deployed in real-world intrusion attempts against US aerospace targets, navigating from a simple IP address to operational objectives automatically.
  • PentAGI: An autonomous offensive framework designed to dramatically lower technical barriers for entry-level hackers. GTIG tracked an attacker using a customized version of PentAGI between late August and early October 2026 to systematically scan and target organizations across Asia.
  • HexStrike AI and CyberStrikeAI: Harnesses that connect large language models directly to over 100 pre-installed offensive security utilities, allowing AI models to dynamically orchestrate end-to-end attack chains.

Security experts note that these multi-agent setups offer hackers unprecedented operational velocity. However, because current autonomous frameworks prioritize speed over stealth, they tend to generate distinct forensic footprints that help defenders spot and contain them.

Beyond network infiltration, hackers are using autonomous agents to shrink the gap between a software patch release and weaponized exploitation—often referred to as the “n-day” window.

GTIG’s analysis shows that while zero-day exploits saw a modest rise through early 2026, averaging 11 per month before spiking to 22 in August 2026, the automation of n-day patch reverse engineering represents the most immediate threat to enterprise infrastructure.

The report also underscores growing financial and infrastructure abuse tied to generative AI. Cybercriminals are increasingly engaging in “LLMJacking,” stealing API keys and cloud credentials from enterprise environments to sell access to high-performance AI models through underground proxy services.

Want to see more of our stories on Google?

Add iPhone in Canada as a Preferred Source on Google

P.S. Want to keep this site truly independent? Support us by buying us a beer, treating us to a coffee, or shopping through Amazon here. Links in this post are affiliate links, so we earn a tiny commission at no charge to you. Thanks for supporting independent Canadian media!

Subscribe
Notify of
guest
0 Comments
Oldest
Newest Most Voted
0
Would love your thoughts, please comment.x
()
x