Rosetta Intel
Briefings/Daily BriefAI FrontierRansomware
Rosetta Lab ↗Blur Horizon LLC
Daily Brief2026-08-26
Daily Brief·2026-08-26·383 Sources·18 Items

Rosetta Daily · Aug 26, 2026

Scanned 383 items across all categories (36-hour window), selected 18.

Actively Exploited / KEV Catalog

Oracle HTTP Server / WebLogic Proxy Plug-in Improper Access Control (CVE-2026-21962, CVSS 10.0)

CISA added this flaw to the KEV catalog on August 24, confirming active exploitation. An unauthenticated attacker with network access over HTTP can access sensitive data — a maximum-severity bug that warrants immediate exposure review and patching.

Sources: The Hacker News · CISA

Zimbra Collaboration Suite OS Command Injection (CVE-2026-73570)

Already on the KEV catalog; CISA gave federal agencies a 3-day remediation deadline. Separately, at least 270 Zimbra servers have already been compromised in ongoing attacks. Mail collaboration platforms are high-value targets — prioritize internet-facing ZCS instances.

Sources: Bleeping Computer · Bleeping Computer

miniOrange SAML 2.0 SSO Plugin Auth Bypass (CVE-2026-61979, CVSS 8.1, WordPress)

Two unauthenticated auth-bypass bugs disclosed by Patchstack are already being exploited to forge SAML responses and log in as WordPress administrators. Not yet in the KEV catalog — sites running the plugin should update immediately.

Sources: The Hacker News · Bleeping Computer

Critical Vulnerabilities

Red Hat Keycloak Password-Reset Logic Flaw (CVE-2026-18963, CVSS 9.1)

An unauthenticated remote attacker can force a password reset flow to take over any account, including administrators. Red Hat and the Keycloak project have shipped patches.

Sources: The Hacker News

Canonical LXD Path Traversal to Host Root Code Execution (CVE-2026-66897, CVSS 9.9)

LXD validates instance-template target paths against a confined os.Root handle, but then creates the file with an unconfined string path — the mismatch lets a user with container-edit permission, or anyone launching a crafted image, escape directory confinement, overwrite root-owned host files, and achieve host root code execution.

Sources: NVD

Netis NC63 Router Firmware Unauthenticated Root RCE (CVE-2026-76070 / CVE-2026-76071, CVSS 9.3)

Two separate stack-based buffer overflows — an oversized Base64-encoded password to the login handler, and an oversized destHost parameter to the ipFilterList=mod action — both in /bin/netis.cgi. Because the device's Boa web server runs the CGI environment as root, unauthenticated remote attackers get root-level code execution.

Sources: NVD (CVE-2026-76070) · NVD (CVE-2026-76071)

DrayTek VigorAP / VigorSwitch Pre-Auth Command Injection (CVE-2026-71914 / CVE-2026-71921, CVSS 9.3)

Multiple VigorAP models fail to validate UDP speed-test messages in the dray_apm component; VigorSwitch models fail to filter the pass field in the setget.cgi interface. Both let a remote attacker execute arbitrary commands as root without authentication. DrayTek gear is a long-running favorite target for botnet scanning — confirm exposure and patch.

Sources: NVD (CVE-2026-71914) · NVD (CVE-2026-71921)

rConfig Auth Bypass to Admin Account Takeover (CVE-2026-77915, CVSS 9.3)

A duplicate Auth::routes() call in routes/web.php re-enables the POST /register route that was meant to stay disabled, and new accounts default to the Admin role. Unauthenticated attackers can self-register a full-privilege admin account and reach stored device credentials, user data, and API token issuance.

Sources: NVD

phpIPAM REST API Authentication Bypass (CVE-2026-67602, CVSS 9.3)

phpIPAM before 1.8.2 keys its object cache by lookup value alone, without the column being queried — so an entry written for an app_id lookup can satisfy a later app_code lookup. Attackers can use a plain database row ID as an API token to read, write, and delete all IP address management records, unauthenticated.

Sources: NVD

RansomLook Itself Carries Three Authorization Flaws (CVE-2026-78370 / 78372 / 78387, CVSS 9.2–9.4)

In an ironic twist, the ransomware-leak-site tracker this pipeline itself relies on had several missing authorization checks: an unauthenticated export endpoint could leak data marked private (groups, markets, ransom notes); several views and API endpoints filtered private entities inconsistently; and an /admin/config endpoint let low-privileged accounts rewrite notification, LDAP, and SMTP settings. All three are fixed in the current release.

Sources: NVD (78370) · NVD (78372) · NVD (78387)

Vendor Advisories

Calix GS7 XGS (GS5239XG) Residential Router — Unpatched NAT Bypass

Used by multiple U.S. broadband providers, this router lets a remote unauthenticated attacker create port-forwarding rules that expose internal, NAT-protected devices to the public internet. No patch and no CVE assigned yet.

Sources: Bleeping Computer

Web Security Research

Trail of Bits: Logic Flaw in Provenance Blockchain's Marker-Account Authorization

Researchers found that the AddAccess check in Provenance (a Cosmos SDK-based public chain) let any user grant themselves admin control over marker accounts without holding any tokens, affecting 82 markers tied to live financial assets on mainnet. Fixed between May and June this year.

Sources: Trail of Bits

PortSwigger: HTML Tag Names Accept a Wider Character Set Than Expected

A close look at what characters browsers actually accept in an HTML tag name — the kind of parsing edge case that can be shaped into malformed markup slipping past common filters.

Sources: PortSwigger Research

AI Security

NVIDIA NemoClaw Prompt-Injection Risk Lets a Malicious Webpage Hijack a Local AI Agent

Oasis Security disclosed that an attacker-controlled webpage could take unauthenticated control of the local Ollama instance backing NemoClaw and plant hidden instructions inside the model — effectively a remote prompt injection against a locally deployed AI agent. Reported to NVIDIA PSIRT.

Sources: The Hacker News

Marimo Notebook Flaw Lets a Crafted Notebook Run Arbitrary MCP Commands

Per a VulnCheck CNA record, opening a specially crafted Marimo notebook in edit mode could execute an attacker-supplied Model Context Protocol command as a local subprocess. Patched.

Sources: The Hacker News

UAT-10147 Uses AI to Scale Server Attacks, Deploying EDR-Evading SPECTRE and a Linux Rootkit

Researchers detail a Chinese-speaking crew, UAT-10147, using AI to scale compromise of Windows and Linux web servers worldwide, concentrated in education, media, technology, and gaming, with victims mostly in Brazil, Bolivia, China, Canada, and Vietnam.

Sources: The Hacker News

Other

North Korea-Linked Actors Poison a Rust Crate With 245 Million Downloads

According to 安全客 (a Chinese-language source), a Rust crate with over 245 million downloads was poisoned with malicious code that auto-executes at compile time, attributed to a North Korea-linked group — another supply-chain poisoning incident. (The source gives no further detail; the specific crate name and any CVE are not confirmed here.)

Sources: 安全客

24 npm Packages Abuse unpkg Mirrors to Host Fake Cloudflare CAPTCHA Pages

A cluster of 24 npm packages is being used as free phishing infrastructure, redirecting victims to ClickFix-style fake CAPTCHA pages. The payload itself is just a single HTML page — installing the package causes no direct harm; npm is simply being abused as free static hosting.

Sources: The Hacker News

← Prev
Rosetta Daily · Aug 25, 2026
Next →
Rosetta Daily · Aug 27, 2026