cybr.cx Daily Digest — August 09, 2026

Critical Vulnerabilities

⚠️ Actively exploited — CVE-2026-8037 | Progress LoadMaster | No CVSS published yet
An unauthenticated command injection vulnerability in Progress LoadMaster allows remote attackers to execute arbitrary commands on the appliance by passing unsanitized input to multiple command endpoints. CISA added this to the KEV catalogue on August 7th with a remediation deadline of August 10th — tomorrow. If you're running LoadMaster at the network edge, treat this as a drop-everything patch event. No authentication barrier means internet-exposed appliances are trivially compromised.

⚠️ Actively exploited — CVE-2026-63077 | JetBrains TeamCity
A deserialization of untrusted data flaw in TeamCity's agent polling protocol allows unauthenticated remote code execution. TeamCity's prevalence in CI/CD pipelines makes this a high-value target for supply chain intrusion — threat actors with a foothold here can poison builds, exfiltrate secrets, and pivot into production environments. CISA's remediation deadline has already passed (August 8th); if you haven't patched, assume exposure.

⚠️ Actively exploited — CVE-2026-18556 & CVE-2026-18577 | N-able N-central
Two authentication bypass vulnerabilities in N-able's N-central RMM platform are being actively exploited — the second (CVE-2026-18577) is an incomplete patch for the first, meaning attackers adapted quickly when the initial fix shipped. RMM tools are a preferred initial access vector for ransomware operators because they provide authenticated, trusted access across entire managed environments. MSPs running N-central should patch both CVEs immediately and audit recent remote sessions for anomalous activity.

⚠️ Actively exploited — CVE-2026-34486 | Apache Tomcat
A missing encryption vulnerability in Tomcat's EncryptInterceptor can be chained with the previously documented CVE-2025-24813 to escalate impact. CISA's remediation deadline has passed. Any Tomcat deployment relying on EncryptInterceptor for session confidentiality should be treated as potentially exposed — check your cluster configs and verify patching status across all nodes.

⚠️ Actively exploited — CVE-2026-9198 | IBM Langflow
A code injection vulnerability in Langflow enables unauthenticated attackers to achieve full remote code execution on default deployments. AI/ML infrastructure has become an increasingly attractive target, and Langflow's default-open configurations have been bitten hard here. If you're running Langflow in any environment — dev, staging, or production — take it off the public internet immediately and apply available mitigations.

CVE-2026-42170 | GIMP | CVSS 7.8 — HIGH
A heap-based buffer overflow in GIMP's DDS file parser triggers when a crafted file declares a D3D9 pixel format with a mismatched bits-per-pixel value. The undersized heap allocation is then overwritten during pixel data processing, potentially leading to arbitrary code execution. The realistic attack vector is a malicious DDS file delivered via phishing or embedded in a project archive — relevant wherever GIMP is used in content production pipelines.

CVE-2026-67620 | Flowise (≤3.1.4) | CVSS 7.7 — HIGH
An SSRF bypass in Flowise's httpSecurity.ts omits Oracle Cloud Infrastructure (192.0.0.192) and Alibaba Cloud (100.100.100.200) metadata endpoints from its deny list. Authenticated attackers can force the server to issue GET requests to these endpoints, potentially harvesting cloud credentials and instance metadata. Cloud-hosted Flowise deployments on OCI or Alibaba are directly at risk; review your deployment's SSRF controls regardless of cloud provider.

CVE-2026-19263 | INQUIRELAB mcp-bridge-api | CVSS 7.3 — HIGH
A command injection flaw in mcp-bridge.js allows remote attackers to manipulate the command/args parameters to execute arbitrary OS commands. The rolling-release nature of this project means there's no fixed version number to target — check your commit hash against b30a82aa and beyond, and avoid exposing any MCP bridge instances to untrusted networks.

Headline News

Metabase SQL Injection Zero-Day Exploited in Customer Data Theft
A critical SQL injection vulnerability in Metabase was weaponised as a zero-day before a patch was available, with attackers breaching customer instances at companies including Framework and Tally to steal user data. The flaw allowed unauthenticated attackers to query underlying databases directly through Metabase's query interface — a particularly damaging vector given that BI tools are typically connected to production databases containing sensitive business and customer records. Organisations self-hosting Metabase should audit their instances immediately, review database access logs for anomalous query patterns, and assess whether downstream data exposure requires breach notification. This incident reinforces the risk of connecting analytics tooling with broad database privileges — least-privilege database accounts for BI tools aren't optional hygiene.

Nearly 800 Malicious npm Packages Deploying Cross-Platform RAT via README Instructions
A campaign distributing close to 800 malicious npm packages has been uncovered, with a twist: instead of using npm's standard lifecycle hooks (preinstall, postinstall) that security tooling commonly monitors, the packages embed malware delivery instructions directly in their README files, triggered when developers call require() in their code. The payload is a cross-platform remote access trojan and infostealer targeting Windows, macOS, and Linux simultaneously — making this a broad-spectrum supply chain threat rather than a targeted OS-specific campaign. The social engineering is elegant and dangerous: developers following README instructions are performing an expected, trusted action, bypassing the suspicion that unusual install-time scripts might trigger. Teams should audit their node_modules dependency trees, enforce lockfiles, and consider tools that inspect package behaviour at runtime rather than just at install time.

Hardware Backdoors Discovered in Some x86 CPUs
Researchers have disclosed the existence of hidden backdoor mechanisms in certain x86 processors — accessible via undocumented instructions that can bypass normal privilege boundaries and grant elevated access to an attacker with the right knowledge. The research, surfacing under the "rosenbridge" project, documents how these covert instruction channels exist beneath the operating system's visibility, meaning traditional software-layer defences are effectively blind to exploitation. The implications for supply chain trust are significant: if hardware itself contains hidden functionality that wasn't disclosed to customers or auditors, the entire security model built on top of that silicon is potentially undermined. Practitioners working in high-assurance or air-gapped environments should treat this as a threat modelling input and watch for firmware or microcode updates from affected vendors.

Schrödinger's Feed

Every major institution has a timeline for when a cryptographically relevant quantum computer will arrive — and almost none of them agree. That dissonance is itself a security signal: the uncertainty range spans years to decades, which makes it genuinely difficult for organisations to decide when to begin PQC migration in earnest. What's not uncertain is that the engineering progress is real — this week a UCLA-led consortium secured $4 million in NSF funding to build a 60 logical qubit trapped-ion architecture, the kind of hardware milestone that steadily shrinks the gap between "theoretical threat" and "operational capability." Practitioners should treat PQC migration planning as a current-year priority, not a future-year problem — especially for data with long-term confidentiality requirements that could be harvested now and decrypted later.

/dev/random

Denmark has decided that if AI can write the essay, students will just have to talk about it instead — the country is now requiring oral defences for written academic work as a systematic countermeasure to AI-assisted cheating. The logic is admirably direct: a language model can produce a flawless essay on Byzantine tax policy, but it cannot sit in a chair and answer follow-up questions from a professor. It's a rare case of an institutional security control that's both low-tech and genuinely hard to circumvent at scale. Somewhere, a student is furiously studying their own AI-generated coursework for the first time.