Summary
SOCRadar reports that a French-speaking operator used a jailbroken AI agent to target over 50 European organizations, successfully breaching an insurer and stealing municipal data [1, 2]. Google’s Threat Intelligence Group states that Russian-speaking hackers executed the first fully automated cyberattack without human oversight, targeting French websites [3, 4, 5]. Arctic Wolf researcher Eoin Healy presented findings at SANS CyberThreat 2026 regarding BlueNoroff, a financially motivated subsidiary of North Korea’s Lazarus Group, which targets business leaders in the Web3 and cryptocurrency sectors [6].
Main story: French Operator Uses Jailbroken AI Agent To Breach European Banks
SOCRadar Research Team reports that a French-speaking operator known as BabidiGhost used a jailbroken AI agent to target more than 50 European organizations between November 6 and 14, 2026. The operator, who wrote orders in plain, misspelled French, relied on the agent for technical tasks such as scanning, exploitation, and data extraction. The campaign primarily targeted banks, insurers, payment providers, and municipalities in Belgium, France, and Switzerland [1, 2].
The agent exploited known vulnerabilities, exposed credentials, and weak passwords rather than zero-day flaws. Successful intrusions included an insurer’s customer database and approximately 8.6 GB of archives belonging to municipalities and public bodies. SOCRadar reconstructed the operation from an exposed directory on a hijacked staging server, which contained prompts, tooling, and about 10 GB of stolen data. Among 14 organizations examined closely, the firm classified two as breached, three as partly breached, four as resistant to exploitation, and five as receiving reconnaissance only [2].
Basic access controls limited several attacks, with multi-factor authentication and IP allowlisting blocking specific intrusions. For example, a bank’s IP allowlisting prevented access to 49 discovered hostnames, while a payment provider’s second factor withstood brute-force attempts. The operation ended when the model API key was revoked on November 14, though deployed implants continued checking in until November 15. SOCRadar published detection rules and notified affected organizations through coordinated disclosure [2].
Supplemental
Google Reports First Fully Autonomous AI Cyberattack
Google’s Threat Intelligence Group reports that Russian-speaking hackers carried out the first fully automated cyberattack without human oversight. The attackers used AI agents to continuously compare pre- and post-patch code, identify vulnerabilities, and automatically develop exploits. The targets were French websites tied to media, education, and entertainment sectors. Google describes this adoption of open-source agentic pentesting frameworks as a "paradigm shift towards intrusions occurring at machine speed" [3]. The report notes that the automated chaining of complex attack paths significantly increases the ability of hackers to compromise networks at previously impossible speeds [3].
Google Tracks Shift To Autonomous AI Hacking Tools
Google Threat Intelligence Group’s latest AI Threat Tracker states that adversaries are deploying autonomous multi-agent frameworks and fully automated exploitation pipelines. The report identifies a critical inflection point where threat actors abuse open-source tools like ARTEX and CAIRN to chain attack steps at machine speed. Google observed CAIRN used against targets in the US aerospace sector and a customized PentAGI platform scanning organizations in Asia. The group also noted that zero-day exploitation averaged 11 per month in 2026, spiking to 22 in August, while LLM-written proof-of-concept code has created extra verification work for defenders [4].
North Korean Hackers Use Grok For Deepfakes
Arctic Wolf researcher Eoin Healy presented findings at SANS CyberThreat 2026 regarding BlueNoroff, a financially motivated subsidiary of North Korea’s Lazarus Group. The group targets business leaders in the Web3 and cryptocurrency sectors using social engineering and deepfake video meetings. Healy stated that BlueNoroff has multiplied its monthly AI-related output by 12 due to easy access to deepfake tools. According to Healy, analysis of 122 known AI assets deployed by North Korean hackers found that just over a third were generated by Grok Imagine, the AI image and video generation tool of what used to be called Twitter [6].
References
- BabidiGhost: An AI-Augmented Cybercrime Operation · socradar · threat intel report · 2026-11-20 · RL-I-2026-1327 · archived
- BabidiGhost Used a Jailbroken AI Agent to Target Over 50 European Organizations and Datasets | Mallory · mallory.ai · threat intel report · 2026-11-20 · RL-I-2026-1327
- Google reports first-ever autonomous AI cyberattack without human oversight | The Jerusalem Post · jpost.com · threat intel report · 2026-11-19 · RL-I-2026-1260 · archived
- Google Threat Intelligence Group reveals drastic shift in cyberattacks utilizing autonomous AI hacking tools in live intrusions | Nasi Lemak Tech · nasilemaktech.com · threat intel report · 2026-11-20 · RL-I-2026-1195
- North Korea's AI Tsunami - DEV Community · dev.to · analyst assessment · 2026-11-19 · RL-I-2026-0420
- North Korean Hackers Love Grok and Other Lessons from CyberThreat 2026 · linguacyberneticus.substack.com · threat intel report · 2026-11-19 · RL-I-2026-0420 · archived