Google reported that China-linked hackers are now running artificial intelligence directly on stolen networks inside the United States to hide their tracks and speed up cyber intrusions.
Story Snapshot
- Google’s threat team says China-linked groups deploy AI agents on victim systems to evade defenses.
- Targets include academic, medical, and military research institutions in North America.
- Attackers blend into normal cloud traffic and use valid accounts to avoid alarms.
- Google has disrupted related infrastructure in past operations, but the tactic is spreading.
Google’s Finding: AI Moves Inside the Victim’s House
Google’s Threat Intelligence Group reported that several China-linked hacking crews have shifted from simple chatbot prompting to using autonomous AI agents that run on compromised networks.
These agents help plan attacks, troubleshoot errors, and route traffic through trusted systems to stay hidden, according to summaries of Google’s latest report.
Running models on already stolen infrastructure lets attackers blend in with normal activity and cut the number of risky connections that might tip off defenders. The move turns every hijacked server into both a launchpad and a smokescreen.
One China-linked group tracked since 2023 focused on academic, medical, and military research organizations in the United States and Canada, with a clear interest in proprietary artificial intelligence research and sensitive data, Reuters reported, citing Google’s findings.
The campaign ran for more than a year before exposure, showing patient tradecraft and a priority on staying quiet long enough to move laterally and siphon valuable information. This is espionage, not smash-and-grab crime. The prize is research that can tilt economic and military advantage.
Why This Matters: Stealth, Speed, and Scale
AI agents can scan for weak points, write and fix code, summarize stolen data, and decide next steps in near real time. Google’s research has warned that state-backed actors, including those tied to the People’s Republic of China, now use artificial intelligence to accelerate every phase of an operation, from reconnaissance to exfiltration.
Another Google analysis noted adversaries increasingly hide inside cloud services, use valid accounts, and even probe artificial intelligence supply chains to get in. This new on-target AI use shrinks the window defenders have to spot unusual behavior.
Past evidence shows these groups adapt quickly when blocked. Google has publicly disrupted China-linked infrastructure controlling attacks across dozens of countries, only to see operators retool and return.
Longstanding reporting tied major Chinese cyber units to broad theft of commercial and defense secrets, reinforcing the strategic motive.
Today’s twist is method, not motive: living off the land now includes living off the victim’s compute, with artificial intelligence doing the heavy lifting behind the scenes.
What Defenders Should Do Now
Security teams should assume intruders will use local compute and trusted cloud routes to hide. Tighten identity controls, watch for strange service-to-service calls, and log model executions on any server that runs artificial intelligence frameworks.
Google’s threat notes describe attackers who rotate internet addresses, automate vulnerability scans, and route traffic through legitimate but compromised environments to avoid alarms.
That demands better baselining of normal traffic, stronger multi-factor authentication, and rapid isolation of any system that shows unexpected model activity or sudden data compression and transfer.
Chinese hackers are running AI on stolen networks to avoid detection, Google says
One China-linked group targets academic, medical and military AI research, the report sayshttps://t.co/ZRMtKlIwML— Emily Turrettini (@textually) September 8, 2026
Leadership should treat this as an economic security issue, not only an information technology problem. The targets—universities, hospitals, and defense researchers—sit at the core of American innovation and national strength.
Congress and agencies should support faster information sharing and tougher penalties for state-enabled theft, while pushing vendors to ship secure defaults that don’t require a hundred manual tweaks to be safe.
The Stakes: A Race We Cannot Afford to Lose
AI-enhanced espionage raises costs for defenders and lowers them for thieves. Google has already flagged attempts to use artificial intelligence to help find unknown flaws and to create adaptive malware.
If attackers keep running models inside victim networks, alarms that once caught odd outbound traffic or foreign command links will miss the action.
That puts the burden on strong identity, tight cloud hygiene, and the will to hunt inside our own walls. Prudence, not panic, wins here—and speed matters.
Sources:
nbcnews.com, reuters.com, services.google.com, cloud.google.com, blog.google, bleepingcomputer.com, euronews.com
