GTIG says exploit_agent wrote and ran exploits with no human in the loop
November 17: CSO Online reports GTIG and Mandiant did not establish that exploit_agent’s autonomous operation caused the data theft found on the staging server.
GTIG says exploit_agent autonomously exploited patched web apps; CSO reports the related data theft is unproven.
Google Threat Intelligence Group and Mandiant said November 16 that a suspected Russian-speaking operator’s exploit_agent compared patches, wrote exploits, and ran them with no human in the loop. An exposed server held stolen data and scripts for reconnaissance, brute force, and webshells; telemetry pointed to French media, entertainment, and education sites. CSO Online reports Google did not establish that autonomous operation caused the theft. Google named no group or state; CyberScoop ties the operator only to Russian-language comments. The tracker also describes in-the-wild CAIRN, PentAGI, and ARTEX. The Register links PentAGI to aerospace, while Google and BleepingComputer attribute that intrusion to CAIRN. Security Affairs says August’s 22 exploited zero-days rose from about eight a month in 2025; CyberScoop says the earlier 2026 baseline was 11 a month.
- Exploit_agent used Russian-language tooling plus frontier and local models.
- Google named no group or state behind the operation.
- FBI FLASH on November 13 said ARTEX use dates to at least August 2026.
- CyberStrikeAI and HexStrike AI were tested, not confirmed in intrusions.
- A modified PentAGI scanned organizations across Asia from late August to early October.
Timeline
dated events in the reporting · oldest first- Nov 13, 2026FBI FLASH says ARTEX use dates to August · Help Net Security
- Nov 16, 2026GTIG and Mandiant publish the AI threat tracker · Cyber Security News
What each outlet adds
beyond the shared facts- CSO OnlineSays autonomous operation was not proven to cause the theft
- The RegisterLinks PentAGI, not CAIRN, to the aerospace sector
- BleepingComputerProjects low-interaction attacks within three months
- CyberScoopPuts the 2026 zero-day baseline at 11 a month
- Security AffairsSays monthly exploited zero-days rose from about eight in 2025
- Threat actors
- MIDNIGHT NEPTUNEStorm-0940UNC6836Midnight Neptune
- AI models
- Gemini
Coverage
oldest first · one row per article · UTCGoogle: AI agent turned patches into exploits and stole data
GTIG says a Russian-language framework autonomously weaponized patches, then stole data from French media, entertainment, and education sites.
GTIG says attackers now run multi-agent AI tools in live intrusions
CAIRN was seen against US aerospace targets; actors also resell stolen LLM access as LLMJacking.
Google says threat groups now automate attacks with AI pentest agents
GTIG says exploit_agent ran without a human and projects low-interaction attacks within three months.
Google: attackers use AI agents to weaponize patches and run intrusions
GTIG says open agent frameworks and LLMs shrink the gap from patch disclosure to n-day exploitation, including PaperCut NG/MF.
Google says AI agent turned patches into exploits with no human help
GTIG says a suspected Russian-speaking operator’s exploitagent framework then stole data from victim organizations.
Google: attackers use AI agents to automate real intrusions
GTIG says CAIRN hit US aerospace targets and a PentAGI variant scanned Asian organizations from late August to early October 2026.
Google GTIG says attackers automated exploit development end to end
An exposed exploit_agent compared code before and after patches and ran exploits; configuration notes were in Russian.
Google GTIG says AI agent frameworks now run multi-stage intrusions
CAIRN and PentAGI were used in real attacks, including against US aerospace, while exploit_agent built and ran exploits with no human in the loop.
Google: Russian-speaking hackers ran first fully automated AI attack
An “exploit_agent” pipeline compared patched code, drafted exploits and ran them with no human in the loop, Google says.
Google says attackers now run autonomous AI exploit agents
GTIG found an exposed multi-agent framework that drafts and runs exploits, plus resale of stolen model access.
GTIG says AI agents wrote and launched exploits with no human
An exposed server ran exploit_agent against patched web apps; an FBI alert says ARTEX has hit multiple sectors since August 2026.
Google expects fully autonomous agentic attacks within three to six months
GTIG says frameworks such as ARTEX already chain known weaknesses, while current attacks still leave obvious forensic traces.
Google finds attacker AI framework that finds and runs exploits
GTIG and Mandiant say exploit_agent targets n-days and was found with victim data; theft attribution is unproven.
Vulnerabilities in this story
All| CVE | Vulnerability | CVSS | EPSS | Signals | Published |
|---|---|---|---|---|---|
CVE-2026-109618+1 related CVE | OOM longjmp memory flaw enabling RCE in PHP | 9.2 | 100% | KEVPoC ×25 | KEV |
| CVE-2026-40217 | LiteLLM through 2026-04-08 allows remote attackers to execute arbitrary code via bytecode rewriting at the /guardrails/test_custom_code URI | 8.8 | 17% | PoC | |
CVE-2026-82078+1 related CVE | Unsafe Reflection RCE in PaperCut NG/MF, Chained with Auth Bypass in Attacks | 9.4 group max | 63% | KEV | KEV |