TLDR INFORMATION SECURITY 2026-11-05

REGISTER FREE — NOVEMBER 12 AT 1PM ET

Rahul Parwani, Product Leader, AI Governance and Security at Airia, and Herman Errico, creator of AARM at Vanta, are hosting a 35-minute fireside on governing AI agents at the point of action. No slides. No demos. You will walk away knowing: * What AARM requires and why it was built * How to evaluate agent security tools in your stack * How to think about build vs. buy at the action boundary

SAVE YOUR SEAT — JOIN LIVE OR GET THE RECORDING

CYBERATTACK ON FRENCH PLASMA STORES IS ENOUGH TO MAKE YOUR BLOOD BOIL

Attackers compromised a refrigeration supplier and altered cold-room settings at France's Blood Establishment, changing temperatures from -30°C to 110°C. EFS monitoring detected the changes. Staff controlled or isolated affected units and moved plasma and samples elsewhere. No blood products were lost. The incident follows European attacks on supermarket refrigeration, including defrost cycles that spoiled food. CERT-FR had urged remediation of vulnerabilities in Danfoss AK-SM and AK-SC controllers.

PHILIPPINE MIGRANT WORKER DATA LIKELY LEAKED IN 69GB BREACH

An unknown actor copied 27 SQL Server backup databases belonging to the Philippine Overseas Employment Administration in 24 minutes. The 69GB set includes migrant-worker records, contracts, accreditation data, and cashier and LandBank financial records. The attacker used GodPotato, SeImpersonate, and Impacket, then left logs and tools on an open Singapore staging server. The stolen backups have not appeared for sale, and the final destination remains unknown. Hunt.io notified CERT-PH and the Department of Migrant Workers.

MSECURE PASSWORD MANAGER WARNS USERS TO SECURE THEIR CREDENTIALS

mSeven notified users of its mSecure password manager that it detected suspicious activity related to the app. Per the message, mSeven warned users to secure their credentials, raising suspicions that it may not be encrypting credentials securely. In an audit conducted last year by Germany's Federal Office for Information Security, the password manager was flagged as not meeting the usual expectations of password managers.

YOU HAD ME AT “HELLO”: INSIDE THE GREETER, A CHINESE‑LANGUAGE FRAMEWORK TARGETING AI INFRASTRUCTURE

Era researchers tracked a Chinese-language framework, Greeter, exploiting Flowise's overrideConfig code-injection flaw (CVE-2024-58351) across seven honeypots from July to October. Each “hello” request delivered a different command through javascriptFunction, without changing the saved chatflow. The framework ran host reconnaissance, searched .env files, SSH keys, and Flowise encryption keys, probed AWS, Azure, and Google Cloud metadata services, and checked Docker socket access, Linux capabilities, persistence paths, and memory-based execution. Of 3,277 identifiable internet-exposed Flowise instances, 703 lacked authentication. 46 matched the honeypot's vulnerable conditions.

FINDING 40+ VULNERABILITIES IN HEALTHCARE SOFTWARE SERVING MORE THAN 100 MILLION PATIENTS

An AI-assisted review of OpenEMR produced 42 published security advisories, fixed in version 8.5.0 on 27 October 2026. The issues include three critical findings: local file disclosure through the PDF generator, SQL injection through patient-data field names, and another SQL injection. Other bugs exposed patient records, allowed portal credential replacement, and enabled administrator API impersonation through an impersonation chain that combined SQL injection and file disclosure to forge an OAuth bearer token, bypassing password and MFA checks.

WHY CONVENTIONAL REPUTATION FEEDS MISS SOFTWARE SUPPLY CHAIN ATTACKS

Conventional threat intelligence feeds and multi-engine scanners fail to detect most software supply chain attacks on day zero, exposing organizations to malicious package behavior for months after public disclosure. Research indicates that blind-spot rates for supply chain attacks in these reputation tools range from 70% to 90%. General-purpose security tools remain ineffective against these threats because they are designed to detect inbound perimeter probes rather than malicious package behavior.

GRIM PROMPTER

Grim Prompter is a Claude Code and Codex skill that has a coding agent build a threat model from a repo and cite file-and-line evidence for every trust boundary, threat, and mitigation. It ships 36 architecture-specific workflows, from RAG and agentic AI to OT/ICS and smart contracts, and 29 method prompts covering STRIDE, PASTA, LINDDUN, and MITRE ATLAS. Saved JSON baselines let the agent review a branch diff against the existing model.

SAPPER-V2

sapper-v2 is a multi-agent pentest harness that won the 634-team qualifier of Tencent Cloud's AI Attack-Defense Challenge, capturing 85 flags in eight unattended hours on 1B tokens where the runner-up burned 4.5B. A commander agent reads a shared fact graph and dispatches scouts, generals, and soldiers through tmux, with GLM-5.3 making decisions and the cheaper DeepSeek V4.1 Flash doing the volume work. It works with any agent CLI, including Claude Code and Codex, and the coordination layer is spun out as the general-purpose nvwabase project.

SENTRA (PRODUCT LAUNCH)

Sentra provides cloud-native data security and governance. It finds and classifies data across cloud, SaaS, warehouses, knowledge stores, and file shares, maps access by users and agents, enforces least privilege, and automates remediation.

VERIFYING APPLE REFERENCE IMAGES

The Privacy Preserving Systems Lab released the first publicly available tool to verify Apple Reference Images captured using the new cryptographic authentication standard for the iPhone 18 Pro series. The system utilizes secure camera hardware alongside Apple's Private Cloud Compute infrastructure to cryptographically sign pixel data. The protocol embeds a developer-accessible reference image next to the main photo using JUMBF boxes, SHA-512 hashing ranges, and a composite ML-DSA and RSA signature.

SOUTH KOREA'S LEE CALLS FOR POWERFUL GOVERNMENT-WIDE CYBERSECURITY BODY

South Korean President Lee Jae Myung called for the creation of a powerful government-wide cybersecurity body to coordinate cybersecurity operations and response. In a cabinet meeting, Lee stated that the current system was too reactive and that a proactive system was needed to protect public services and critical infrastructure. Lee also stressed the need for further public- and private-sector collaboration.

SECURING OUR FRONTIER MODEL TRAINING ENVIRONMENT

At the end of September, OpenAI embarked on a frontier model training pause to harden its training infrastructure against potential escapes. Beyond host-based sandboxing, OpenAI limits its model training environment's network access to strictly necessary internal services and an “offline internet” of cached search indexes and pages. OpenAI also spent billions of tokens red-teaming its training environment to find gaps in its hardening.

FRONTLINE EDUCATION DATA BREACH IMPACTS OVER 1.3 MILLION

Hackers exploited a third-party software vulnerability on August 14 to access Frontline Education systems.

CHAINSTACK REPORTS PRODUCTION SYSTEMS BREACH EXPOSING CUSTOMER DATA AND PASSWORD HASHES

Chainstack says an attacker entered via a third-party component on October 26 and reached production systems, accessing customer names, organizational details, email addresses, and password hashes.

RESIDENTIAL PROXIES DETECTED ON WHITE HOUSE NETWORK

Proxy-detection vendor Layer3 Intel says its probes exited through the White House's own network (AS2650) 236 times in a month via 6 addresses and 28 commercial residential proxy services.