cybr.cx Daily Digest — September 06, 2026

Critical Vulnerabilities

⚠️ Actively exploited — CVE-2026-85046 | Google Chromium V8 | CVSS: Not scored (KEV)
A type confusion vulnerability in Chromium's V8 JavaScript engine allows a remote attacker to execute arbitrary code inside the browser sandbox via a crafted HTML page. Every Chromium-based browser is affected — Chrome, Edge, Brave, and others. CISA added this to KEV on September 4th with a patch deadline of September 18th, and exploitation is confirmed in the wild. Update your browsers now; don't wait for managed deployment cycles.

⚠️ Actively exploited — CVE-2026-82329 | JFrog Artifactory | CVSS: Not scored (KEV)
Under its default configuration, Artifactory allows an unauthenticated attacker with network access to obtain administrative privileges. This is as bad as it sounds — full admin takeover without credentials on what is frequently a crown-jewel artifact repository. Patch immediately and audit access logs for anomalous admin-level activity since early September.

⚠️ Actively exploited — CVE-2026-83548 & CVE-2026-83549 | SonicWall SMA1000 | CVSS: Not scored (KEV)
Two chained vulnerabilities in SonicWall SMA1000 appliances: an unauthenticated SSRF (CVE-2026-83548) for initial access and an authenticated OS command injection (CVE-2026-83549) for RCE. Together they form a full pre-to-post-auth exploit chain against a perimeter device. Both have CISA remediation deadlines of September 5th — if you haven't patched yet, assume compromise and investigate.

⚠️ Actively exploited — CVE-2026-49869 | Kestra OSS | CVSS: Not scored (KEV)
Kestra's workflow orchestration platform contains an unauthenticated OS command injection flaw allowing remote attackers to create and execute arbitrary workflows without credentials. Workflow orchestrators have privileged access to infrastructure, making this exceptionally dangerous. CISA's remediation deadline was September 5th.

⚠️ Actively exploited — CVE-2026-82078 & CVE-2026-81578 | PaperCut NG/MF | CVSS: Not scored (KEV)
A pair of chained PaperCut vulnerabilities: CVE-2026-81578 allows unauthenticated configuration modification, which can be chained with CVE-2026-82078's unsafe reflection flaw to execute arbitrary Java bytecode under the PaperCut server process context. PaperCut has been a persistent target for ransomware operators — treat these as critical and patch now.

⚠️ Actively exploited — CVE-2026-9586 | Sangoma Switchvox | CVSS: Not scored (KEV)
A single crafted HTTP request is sufficient for an unauthenticated attacker to execute arbitrary SQL against the backend PostgreSQL database in Sangoma Switchvox, with a path to full RCE. VoIP infrastructure is a high-value target for toll fraud and eavesdropping operations. Patch deadline was September 5th.

⚠️ Actively exploited — CVE-2026-59822 | BerriAI LiteLLM | CVSS: Not scored (KEV)
LiteLLM's MCP Streamable HTTP endpoint accepts arbitrary Bearer tokens, allowing unauthenticated attackers to establish fully authenticated MCP sessions. As LiteLLM is widely deployed as an AI gateway in enterprise environments, this gives attackers access to backend LLM infrastructure and any connected tooling.

⚠️ Actively exploited — CVE-2026-48710 | Kludex Starlette | CVSS: Not scored (KEV)
HTTP request smuggling in Starlette allows attackers to inject paths into the host header, prepending them to the actual request path. The practical impact is authentication bypass in any application that reconstructs URLs from the host header for access control decisions — a pattern that's more common than it should be.

CVE-2026-86169 | Axolotl (ML training framework) | CVSS: 8.8 | HIGH
A misconfigured default in Axolotl's multipack patch path sets trust_remote_code to None rather than False, bypassing the security guard entirely. Attackers who publish a malicious Hugging Face model repository and get a victim to select it as a base model achieve arbitrary Python code execution on the training host. Anyone using Axolotl for fine-tuning should verify base model provenance and update to 0.18.1+.

CVE-2026-86123 | SQL Chat | CVSS: 8.7 | HIGH
Four unauthenticated API endpoints in SQL Chat accept client-supplied database connection parameters and execute arbitrary SQL queries against attacker-specified hosts. This is essentially a built-in authenticated SSRF and SQL injection combo — anyone running SQL Chat exposed to untrusted networks should treat it as a critical exposure.

CVE-2026-86117 | Coolify | CVSS: 8.1 | HIGH
The OAuth callback handler in Coolify signs users in based purely on email address without verifying the OAuth provider's assertions. An attacker who knows a target's email can register that address with any enabled OAuth provider and immediately gain authenticated access as that user. Self-hosted infrastructure management platforms are high-value targets.

Headline News

Chromium V8 Zero-Day Exploited in the Wild — Update Now

A type confusion flaw in Google's V8 JavaScript engine (CVE-2026-85046) is being actively exploited against Chromium-based browsers including Chrome, Edge, and others. The vulnerability allows a remote attacker to achieve arbitrary code execution within the browser sandbox via a specially crafted HTML page — meaning a drive-by visit to a malicious site is sufficient for compromise. Google's patch, bundled with fixes for eleven other flaws including ten rated high-severity, is already shipping via automatic updates, but enterprise environments with managed update policies may lag behind. Security teams should verify browser versions across their fleet immediately and consider pushing forced updates ahead of schedule — CISA's remediation deadline of September 18th should be treated as an absolute ceiling, not a target.

ASCII Smuggling Jumps from AI Attacks to Phishing Campaigns

Invisible Unicode tag characters — originally studied as a vector for manipulating AI models through hidden prompt injections — have found a new home in conventional phishing campaigns. Attackers are embedding these zero-width characters in URLs and document payloads to bypass content filters and security scanners that process visible text, while the hidden characters alter what the browser or application actually parses and executes. The technique is particularly effective against tools that perform textual analysis of links or email bodies, since the invisible characters are genuinely absent from most rendering pipelines but present in the underlying byte stream. Practitioners should evaluate whether their email security gateways and URL scanners are performing Unicode normalisation and tag-character stripping on inbound content — most deployed configurations currently do not.

Government Rails Application Exploited Hours After CVE Disclosure

A government-operated Ruby on Rails application was successfully attacked within hours of a CVE patch being published for the underlying framework — a stark illustration of how narrow the window between disclosure and exploitation has become for widely-deployed software. Attackers appear to have reverse-engineered the patch to identify the vulnerable code path and weaponised it before administrators had a reasonable opportunity to respond. The incident reinforces a pattern security teams have been observing for years: disclosure timelines that assume days or weeks of breathing room no longer reflect operational reality. Practitioners maintaining Rails applications — particularly those in the public sector where update cycles tend to be slower — should review patch management processes to prioritise framework-level CVEs for same-day or next-day deployment where possible.

Schrödinger's Feed

Europe is moving to formalise its quantum ambitions through a proposed Quantum Act, designed to convert the continent's considerable research output into industrial and commercial capability — with security infrastructure implications baked into the policy conversation. The framing is notable: the bottleneck is no longer physics, but engineering, suggesting that cryptographically relevant quantum hardware is an engineering milestone rather than a theoretical one. For practitioners, this signals that government procurement and infrastructure planning in the EU will increasingly treat post-quantum cryptography migration as a near-term compliance obligation rather than a distant concern. If you haven't mapped your organisation's exposure to RSA and ECC-based systems yet, a policy framework that accelerates quantum hardware development is a reasonable forcing function to start.

/dev/random

This week's deep dive into Rust's dyn Trait vtable memory layout is doing the rounds among systems programmers — and it's genuinely illuminating for anyone who has ever wondered what the compiler is quietly constructing on their behalf. When you use dynamic dispatch in Rust, the compiler builds a fat pointer: one pointer to the data, and one to a vtable — a static table of function pointers for each trait method, plus size and alignment metadata. Unlike C++ vtables embedded in object headers, Rust's are stored separately, which means the same piece of data can be viewed through different trait lenses at the cost of an extra pointer dereference. It's a reminder that "safe" doesn't mean "magic" — there's a very concrete memory layout under the abstraction, and understanding it matters for anyone writing unsafe code or auditing for memory safety issues at the interface between Rust and FFI.