cybr.cx Daily Digest — August 30, 2026

Critical Vulnerabilities

⚠️ Actively exploited — CVE-2023-49105 | ownCloud | CVSS 9.8 (Critical)
CISA added this to the KEV catalogue on August 27 with a patch deadline of today. The flaw allows unauthenticated access to any file if the target username is known and the account lacks a signing-key configured — which is many default deployments. If you're running ownCloud, treat this as a fire drill: patch, rotate credentials, and audit access logs immediately.

⚠️ Actively exploited — CVE-2026-53362 | Linux Kernel | CVSS TBD
A privilege escalation vulnerability in the IPv6 networking subsystem is being exploited in the wild, affecting distributions including SUSE, Red Hat, and any other Linux-based product. CISA's due date is also today. Scope is broad — this touches nearly every Linux workload. Apply kernel patches from your distro vendor now; where patching is not immediately possible, consider disabling IPv6 at the kernel level as a temporary mitigation.

⚠️ Actively exploited — CVE-2026-66384 | JFrog Artifactory | CVSS HIGH
An authenticated path traversal vulnerability allows data to be written outside the intended Docker cache path under specific remote-repository configurations. Active exploitation means attackers with even low-privilege credentials can potentially stage malicious artifacts. Patch due September 10, but don't wait — audit remote repo configurations and restrict write permissions now.

⚠️ Actively exploited — CVE-2021-23758 | Ajax.NET Professional (AjaxPro) | CVSS HIGH
A .NET deserialization vulnerability in this older library is seeing fresh exploitation — a reminder that threat actors routinely revisit ancient bugs against organisations running end-of-life software. CISA recommends discontinuing use entirely. Audit your web estate for AjaxPro dependencies; there is no supported patch path.

⚠️ Actively exploited — CVE-2026-60004 | Gitea | CVSS HIGH
Any user with repository write access can send a crafted patch to the diffpatch API endpoint to plant an executable Git hook, executing arbitrary shell commands as the Gitea service account. This is a significant supply chain risk in self-hosted Gitea environments. Upgrade immediately and audit recently merged patches for signs of hook manipulation.

⚠️ Actively exploited — CVE-2019-1068 | Microsoft SQL Server | CVSS HIGH
A 2019-vintage SQL Server RCE is back in active exploitation — attackers can execute code in the context of the Database Engine service account. If your SQL Server instances are still unpatched against a seven-year-old CVE, that conversation needs to happen today.

⚠️ Actively exploited — CVE-2022-0995 | Linux Kernel | CVSS HIGH
An out-of-bounds memory write in the Linux kernel allows local users to gain root or cause denial of service. Combined with CVE-2026-53362 above, Linux defenders are facing a two-front kernel fight this weekend. Prioritise kernel patching across your estate.

⚠️ Actively exploited — CVE-2026-8452 | Citrix NetScaler ADC & Gateway | CVSS HIGH
A memory buffer bounds vulnerability leading to denial of service is being actively exploited against Citrix edge devices. The patch due date was yesterday — if NetScaler devices in your environment haven't been updated, they should be considered compromised until verified otherwise.

CVE-2026-82447 | Skyvern < 1.0.45 | CVSS 8.8
A double-render flaw in TextPromptBlock processes Jinja templates first in a sandboxed environment, then again in an unsandboxed one — meaning attackers who control workflow parameters or upstream block output can escape the sandbox entirely and execute code with server process privileges. Upgrade to 1.0.45 or later; this is exploitable by any user who can influence workflow inputs.

CVE-2026-18729 | IBM Langflow OSS 1.0.0–1.11.1 | CVSS 8.8
Authenticated remote code execution via improper control of code generation — one of three Langflow CVEs published today (see also CVE-2026-18891 at 8.2 and CVE-2026-18904 at 8.2, covering unauthenticated flow execution and namespace collision respectively). The full trilogy means Langflow deployments are exposed from authentication bypass through to RCE. Treat any public-facing Langflow instance as compromised pending patching and audit.

CVE-2026-82450 | BookStack < 26.05.4 | CVSS 8.8
Users with Import Content and Create Books permissions can upload a PHP polyglot file disguised as a book cover inside a portable ZIP import. BookStack stores it in the public web root and the server executes it — classic web shell via trusted feature. Upgrade to 26.05.4 and review recent ZIP imports in your audit logs.

CVE-2026-82461 | pac4j-oidc < 6.5.6 | CVSS 8.1
The library fails to verify access token signatures, issuers, audiences, or expiry when extracting Keycloak realm and client roles. An attacker can forge an access token granting administrative roles and pair it with a valid ID token to bypass all role-based authorisation. Any application relying on pac4j for OIDC role validation is affected — upgrade immediately and review whether forged tokens have been submitted against your endpoints.

Headline News

Cosmos EVM Exploited Across Six Blockchains After Known Vulnerability Went Unpatched

A critical balance-handling flaw in the shared Cosmos EVM module — tracked as GHSA-7g4w-cg88-2cq2 — was exploited across six separate blockchains between August 20 and 25, draining funds before the affected chains could respond. What makes this incident particularly damaging is the disclosure timeline: Cosmos Labs was aware that every blockchain running the shared module was vulnerable before exploitation began. The attack is a textbook example of the shared-library risk that plagues modular blockchain ecosystems — a single vulnerable component propagates across every downstream project simultaneously, and the disclosure-to-patch window becomes a race between maintainers and attackers. For security practitioners working in or adjacent to Web3 infrastructure, this reinforces the need to treat upstream dependency advisories with the same urgency as CVEs in traditional software supply chains.

Claude Code Susceptible to Prompt Injection via Webpage Summarisation

A researcher demonstrated that Anthropic's Claude Code — an agentic coding assistant with the ability to execute shell commands and interact with filesystems — can be hijacked through a straightforward prompt injection attack: simply asking it to summarise a malicious webpage is sufficient to redirect its actions. The attack exploits the gap between what a user intends and what the model executes when it encounters adversarially crafted content in the environment it browses. This is a significant concern because Claude Code operates with substantially more system access than a conversational AI — a redirected agentic session can write files, modify code, or exfiltrate data. The finding adds to a growing body of research demonstrating that indirect prompt injection via untrusted external content is not a theoretical edge case but a reliable and repeatable attack class against AI agents with tool access. Teams deploying agentic AI in developer workflows should treat external content retrieved by these tools as untrusted input, applying the same scrutiny as any other data ingested by privileged processes.

Love Electric Breach: 877,000 Driver Records Listed for $600

A threat actor claiming to hold data from UK electric vehicle salary sacrifice provider Love Electric listed 877,000 driver records on an English-language breach forum on August 26, asking just $600 for the full dataset. The alleged data includes sensitive personal and employment information tied to employees who used the scheme — the kind of enriched dataset that enables targeted phishing, identity fraud, and corporate espionage when employment records and personal identifiers are combined. The incident highlights the identity risk created by third-party benefit and salary sacrifice providers, which collect sensitive HR-adjacent data but may not be subject to the same security scrutiny as primary employers or payroll systems. Organisations offering or administering salary sacrifice schemes should be asking hard questions about their providers' security posture and ensuring breach notification obligations flow contractually back to them.

Schrödinger's Feed

IBM has released Qiskit Fermions 0.1.0, an open-source library and transpiler extension that preserves fermionic operator structure when compiling quantum circuits down to qubit architectures — built on a high-performance Rust core with Python bindings. This might sound purely academic, but better transpilation that preserves physical structure means more efficient, less error-prone circuits, which accelerates the timeline to fault-tolerant quantum computation. Fault-tolerant quantum machines are the milestone at which current asymmetric cryptographic assumptions — RSA, ECDH — begin to break down in practice. Practitioners managing long-lived PKI infrastructure or secrets with multi-decade sensitivity windows should treat incremental tooling advances like this as part of the same threat model as nation-state adversaries: slow-moving, but directionally certain.

/dev/random

A developer published "Tether," a project that routes iMessage and SMS through a paired iPhone to a Linux desktop — bridging Apple's walled messaging garden to an OS Apple has never officially acknowledged exists. The technical approach mirrors what various commercial tools have attempted, but the open-source implementation lays bare exactly how much trust you're extending to a local proxy process sitting between your encrypted messages and your screen. It promptly topped Hacker News, which perhaps says more about Linux users' feelings toward Apple's ecosystem than it does about the project's security model. Read the code before you route your iMessages through anything, is the takeaway.