Security news from the BlueTeamSec community for Friday the 11th of September 2026. Quite a busy one today, with fourteen stories to get through.
Microsoft have written up a passkey-themed social engineering campaign targeting enterprise cloud environments. Attackers are impersonating IT support to trick employees into granting access to organisational cloud resources β essentially getting victims to compromise their own identity.
Volexity and Proofpoint both covered the same campaign β two Chinese threat actors, UTA0560 and JungleBamboo, exploited a patch-gap window in Chrome this month, chaining three zero-day vulnerabilities against NGOs. The interesting bit is that both actors used identical shellcode, which suggests a shared exploit supply chain rather than independent development.
And another exploit in the wild β Cisco Talos report active exploitation of two critical vulnerabilities in Cisco Secure Firewall Management Center, one of which is a perfect-ten authentication bypass that gives root access. Three distinct threat actor clusters are using it, including for Qilin ransomware operations.
NetSPI have written up StyleSmuggler, a critical remote code execution flaw in Adobe Commerce and Magento that's been actively exploited since the fourth of September. Attackers are using the email template engine to inject malicious code and deploy Rust-based Linux malware and PHP web shells. CISA added it to the Known Exploited Vulnerabilities catalogue on the eighth.
GreyNoise published an interesting piece on an AI-orchestrated campaign against PaperCut servers. A likely Russian-speaking actor used AI agents from DeepSeek and OpenAI to automate exploitation of PaperCut vulnerabilities, compromising four hundred and forty instances across nearly four hundred organisations in forty-eight countries. The standout detail is speed β domain admin access in as little as seven minutes.
Hunt Intelligence documented a cryptomining botnet that's compromised over three and a half thousand Redis servers by exploiting instances without authentication. The operator abuses the legitimate master-replica replication protocol to inject miners, and helpfully exposed the entire operation through their own misconfigured file server.
TeamT5 have a write-up on a Chinese phishing campaign using the Darcula kit to harvest Taiwanese credit card data. The attackers are impersonating government services and public utilities through smishing and email, with a particular focus on the E-invoice service.
Security researchers have documented a fileless execution technique for Linux that leverages the kernel keyring subsystem to run binaries without filesystem artifacts or standard system calls. It evades the usual monitoring touchpoints β no file descriptors in proc, no disk inodes β which presents fairly significant detection challenges.
Intigriti published research on vulnerabilities in AI customer service agents, demonstrating advanced prompt injection and protocol exploitation techniques. The attack leverages UI gaps between human dashboards and AI processing, using invisible text and MIME manipulation to smuggle instructions that weaponise the agent's ability to execute backend tool calls β password changes, fund transfers, that sort of thing.
Huntress completed a six-week study of Active Directory Rights Management Services and discovered four methods to extract encryption keys, including a non-rotatable private key with a two hundred and fifty-five year lifespan. The extracted key enables offline decryption of every document ever protected by that deployment, without server contact or legitimate user rights, and the vulnerability cannot be remediated through key rotation.
TokenAI released Horus Cyber Nano, a sixteen-billion parameter open-weights AI model designed for defensive security work. It's built on a Mixture-of-Experts architecture and handles secure code review, vulnerability analysis, terminal workflows, and authorised red team reasoning, with support for text, image, and video inputs.
A researcher audited the top one hundred cybersecurity vendors for adoption of the llms.txt standard introduced in September 2024 for guiding AI agents. Only thirty-four of one hundred serve a functional file β the rest show widespread non-compliance including WAF blocks, technical errors, and improperly formatted content.
And finally, a research tool called TATS β Token Analysis and Tracking System β which captures, decodes, and analyses OAuth, OIDC, and Microsoft Entra ID tokens from network traffic sources like Burp Suite. It stores tokens and claims in a SQLite database with a web dashboard, which is useful for research but does create obvious security risks if that database ends up somewhere it shouldn't.
That's your lot for today. Attribution sits with the original authors β the analysis here was automated. We'll be back with the next briefing soon.