READ. SCROLL. LISTEN.

Unbiased headlines. Facts, not spin.

Unbiased Headlines is an independently owned news organization. We bring you the facts, not opinions, written fresh from 110+ named sources. Sources linked so you can verify it yourself.

← Back to headlines

Google Says Hackers Have Automated Exploits With AI Agents, Projects Largely Hands-Off Attacks Within Three Months

Google Says Hackers Have Automated Exploits With AI Agents, Projects Largely Hands-Off Attacks Within Three Months
Google's Threat Intelligence Group said today, Nov. 16, that attackers have moved from experimenting with AI to running agentic frameworks in real intrusions. Google estimates low-sophistication attacks needing little to no human involvement within three months, and sophisticated autonomous ones within six. The defenses it recommends are old ones, but the time defenders have to apply them is shrinking.

Google's Threat Intelligence Group (GTIG) published new findings today, Nov. 16, saying hackers have moved past asking chatbots for help and are now running multi-agent AI frameworks that find and exploit vulnerabilities on their own.

Google says both financially motivated criminals and state-backed groups have adopted the tools at scale.

The company's estimate: within three months, attackers will be able to launch unsophisticated attacks that need "little to no human interaction." Within six months, GTIG says, "sophisticated, fully autonomous" agentic attacks are likely to emerge. Those are Google's projections, not observed events.

What Google says it has already seen

GTIG says that since late August 2026, threat actors have been operating open-source agentic pentesting platforms including ARTEX, PentAGI, HexStrike and CAIRN. Google's incident-response teams and its Mandiant unit saw them used across multiple phases of the attack life cycle. Pentesting tools are built for security professionals, but Google says attackers are now pointing them at victims.

The FBI issued its own advisory last week. It warned that multiple threat actors have weaponized the ARTEX framework since August, hitting organizations in the financial, government, technology, defense and healthcare sectors.

Google also cited CAIRN's use in an attack on the U.S. aerospace sector. It said suspected North Korean hackers used Claude Code to deliver malware, and that suspected China-aligned espionage actors have been experimenting with these frameworks.

The exploit_agent server

The most striking finding came from an attacker's mistake. GTIG found an exposed staging server hosting a multi-agent vulnerability research and weaponization framework it named exploit_agent. The framework plugs into multiple frontier and local large language models.

The server also held stolen victim data, reconnaissance information, webshells and brute-forcing scripts.

Google says the system analyzes software updates on popular web applications, looks for signs of security flaws, and automatically attempts to exploit them. It was used 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 in its report.

Patches become a roadmap

Google also describes attackers using AI to reverse-engineer security patches by comparing code versions, then turning the result into working exploits. It pointed to flaws in PaperCut NG/MF earlier this September as an example.

Google's reasoning is that patched, publicly known ("n-day") bugs give attackers a finite set of material to work from: source code diffs, release notes, security research and proof-of-concept code. That shrinks the gap between a patch landing and active exploitation.

The flood of fake help is a problem too. After PHP announced two flaws, hundreds of purported proof-of-concept exploits appeared, mostly AI-generated, according to GTIG. It said the pile of junk added a significant burden for defenders sorting real threats from noise.

Stolen AI access is its own market

The report also flags "LLMjacking." Attackers steal LLM credentials and cloud resources from compromised enterprise environments, then resell the access through underground proxy services such as APIMart and JetNight, at discounted prices.

Google further described a Chinese-language phishing platform called Jieyou that has integrated an LLM chatbot assistant. Its conclusion is that open-source frameworks have lowered the barrier to entry for newer attackers.

Same weaknesses, faster

Google's own account is less apocalyptic than the headline numbers suggest. It says the mechanics of these intrusions remain the same: compromised credentials and unpatched systems. What changed is speed and scale.

Its recommended defenses are control-plane isolation, short-lived credentials, least privilege, segmentation and behavioral detection. Google itself calls these long-standing security best practices.

Chris Betz, Google Cloud's chief information security officer, made a related argument in the company's 2027 Cloud CISO Perspectives forecast. He wrote that multi-stage attacks now unfold in seconds, leaving weekly vulnerability scans and month-long patch cycles too slow. The forecast says attackers are using multi-agent frameworks to harvest credentials at large scale in under six hours.

The forecast also warns that attackers may shift from stealing data to "decision manipulation," poisoning data or business logic so autonomous systems make harmful choices on things like contracts, inventory routing or capital allocation. It flags software supply chain risk from open-source libraries and MCP tools that coding assistants pull into corporate codebases.

For defense, the forecast recommends treating every AI agent as a first-class identity with verifiable credentials and least-privilege access. It also recommends teams of specialized agents (hunter, decoder, responder) under a central coordinator that escalates to humans, rather than one general-purpose agent. It describes unmanaged "agent sprawl" as a new form of shadow IT.

What to watch

Google is the author of both the findings and the forecast, so the three-month and six-month timelines are its own estimates. Three months from today runs to mid-February 2027, and six months to mid-May. Whether fully autonomous attacks show up on that schedule is the open question. In the meantime, the FBI advisory on ARTEX is already out and the exploit_agent infrastructure has been exposed.

Sources used for this briefing

6 links from 6 independent newsrooms (syndicated copies of the same report count once)

These are the sources it draws on, linked so you can verify.

center
The Hacker NewsGoogle Finds Hackers Using AI Agents to Build and Run Exploits From Software Patches
unknown
BleepingComputerGoogle warns that fully-automated AI attacks are coming soon
unknown
TechBytesGoogle warns that fully-automated AI attacks are coming soon
unknown
securitybrief.com.auGoogle Cloud warns AI agents will reshape cyber defence
unknown
securitybrief.caGoogle Cloud warns AI agents will reshape cyber defence
unknown
cloud.googleARTEX and Friends: Enter Agentic Offensive Frameworks | Google Cloud Blog

More coverage: Google