cybr.cx Daily Digest — May 20, 2026

Critical Vulnerabilities

CVE-2026-5804 | Motorola Factory Test Component | CVSS 8.4 (HIGH)
A pre-installed factory test app (com.motorola.motocit) on Motorola devices exposes a writable file descriptor in external storage, allowing any third-party app on the device to spin up a TCP server and inherit sensitive system permissions. No user interaction is required beyond installing a malicious app. Motorola device fleet owners — particularly enterprise MDM environments — should audit installed system components and apply any available OEM patches immediately.

CVE-2026-7504 | Keycloak (wildcard redirect configs) | CVSS 8.1 (HIGH)
Keycloak's redirect URL validation can be bypassed when clients are configured with wildcard (*) entries in the "Valid Redirect URIs" field — a configuration that's unfortunately common in dev and staging environments that never got hardened. An attacker crafts a malicious request that survives validation and redirects victims to attacker-controlled URLs, enabling credential harvesting or downstream OAuth abuse. Audit your Keycloak client configs now and replace wildcards with explicit URIs.

CVE-2026-8711 | NGINX JavaScript (njs) | CVSS 8.1 (HIGH)
When the js_fetch_proxy directive is configured with client-controlled NGINX variables (e.g., $http_*, $arg_*, $cookie_*) alongside ngx.fetch() calls, an unauthenticated attacker can trigger a heap buffer overflow in the NGINX worker process via crafted HTTP requests. This is a remote code execution risk on affected configurations. Review any njs deployments using dynamic variables in js_fetch_proxy and patch or restrict input surface immediately.

CVE-2026-7307 | Keycloak SAML endpoint | CVSS 7.5 (HIGH)
A remotely exploitable DoS: unauthenticated attackers can send malformed XML to Keycloak's SAML endpoint and cause runaway CPU consumption, starving worker threads until the server stops responding. No authentication required. If you're exposing Keycloak SAML to the internet, this is a real availability risk — patch, rate-limit the SAML endpoint, or place a WAF in front.

CVE-2026-7507 | Keycloak session fixation | CVSS 7.5 (HIGH)
The /login-actions/restart endpoint processes session handles without CSRF protection or cookie ownership validation, enabling session fixation attacks. An attacker pre-creates an auth session, sends the victim a crafted link, and can potentially hijack the resulting authenticated session. This one requires social engineering but is straightforward to exploit once the groundwork is laid. Patch Keycloak and enforce strict session binding policies.

CVE-2026-43634 | HestiaCP 1.2.0–1.9.4 | CVSS 7.5 (HIGH)
HestiaCP blindly trusts the CF-Connecting-IP HTTP header without verifying requests actually originate from Cloudflare's network. An unauthenticated attacker can spoof any IP address in this header, bypassing fail2ban protections and per-user IP allowlists entirely. If you're running HestiaCP — a popular open-source hosting control panel — upgrade to 1.9.5+ or manually validate Cloudflare IPs at the network layer before the header is consumed.

CVE-2026-8912 | Contest Gallery WordPress Plugin ≤28.1.6 | CVSS 7.5 (HIGH)
An unauthenticated SQL injection via the form_input parameter in the post_cg_gallery_form_upload AJAX action. No login required, and the vulnerable endpoint is publicly accessible by default. Standard severity, standard advice: update the plugin, and consider a WAF rule blocking unsanitised input to AJAX endpoints if patching is delayed.

CVE-2026-8073 | Kirki WordPress Plugin ≤6.0.6 | CVSS 7.5 (HIGH)
The downloadZIP function in the Kirki page builder plugin lacks both path validation and capability checks, allowing unauthenticated attackers to read and delete arbitrary files within the WordPress uploads directory. File deletion without authentication is a meaningful defacement and availability risk. Update immediately; the plugin has a large install base.

Headline News

CISA Contractor Exposed AWS GovCloud Credentials on GitHub
A contractor working with CISA — the agency nominally responsible for hardening the U.S. government's cyber posture — left active AWS GovCloud access keys in a public GitHub repository. The exposure was described by one observer as "the worst leak I've witnessed," a remarkable statement in a field accustomed to credential disasters. GovCloud environments are specifically designed to handle controlled unclassified and sensitive government workloads, meaning the blast radius of a key compromise extends beyond typical cloud misconfigurations. The incident is a vivid reminder that supply chain trust doesn't stop at code: contractors with privileged cloud access are part of the attack surface, and secret scanning on public repositories should be a non-negotiable baseline for any organisation handling sensitive infrastructure. Practitioners should treat this as a prompt to audit third-party contributor access to internal repositories and verify that automated secret detection is actually running — not just theoretically enabled.

Mini Shai-Hulud Worm Escalates npm and PyPI Supply Chain Campaign
The TeamPCP supply chain campaign has significantly intensified, with 314 npm packages confirmed compromised in its latest wave — including high-profile packages such as echarts-for-react, size-sensor, and 271 packages under the @antv scope. The campaign also claimed an officially confirmed compromise of the Checkmarx Jenkins plugin, a tool widely used in enterprise CI/CD pipelines, meaning malicious code may have been introduced into build toolchains at scale. The self-spreading "Mini Shai-Hulud" worm component is what makes this campaign operationally unusual: compromised packages actively attempt to spread the infection to co-installed packages in the same environment. For defenders, this means a single compromised dependency could have mutated your broader dependency tree before detection. Security teams should audit their npm and PyPI lockfiles against the published IOC lists and treat any Checkmarx Jenkins plugin instance as potentially tainted until verified clean.

1.8 Million Patient Records Stolen in Healthcare Network Breach
Attackers spent approximately three months — from November 2025 through February 2026 — moving quietly through the systems of the largest public health provider in the United States before exfiltrating records belonging to at least 1.8 million patients. The data accessed includes highly sensitive clinical and personal information. The dwell time here is the critical detail: three months of undetected access in a healthcare environment suggests either absent or overwhelmed detection capabilities, a common reality in underfunded public health IT. Healthcare remains among the highest-value targets for both financially motivated ransomware groups and state-affiliated actors interested in population-level health data. Practitioners in the sector should prioritise anomalous data egress detection and lateral movement monitoring — the initial access question matters less than what happens in the weeks after.

Schrödinger's Feed

The U.S. Emerging Research and Venture Accelerator (ERVA) has published an engineering report on quantum technology readiness, with the co-chair noting a pointed assessment: U.S. quantum science remains world-class, but global competition is accelerating and the key limiter is no longer the underlying physics — it's the engineering, manufacturing, and talent pipeline required to turn research into operational systems. That framing matters for security practitioners, because the gap between "quantum computers that threaten RSA" and "we're still doing science" is increasingly being measured in engineering lead time, not fundamental breakthroughs. Meanwhile, Aramco and Pasqal have stood up the Middle East's first commercial Quantum Computing as a Service platform in Dhahran — a sign that quantum access is regionalising faster than post-quantum cryptography adoption is progressing in many organisations. If your organisation still has a "we'll worry about PQC when quantum is real" posture, the engineering clock is the one to watch now.

/dev/random

OpenAI has adopted Google's SynthID watermarking standard for AI-generated images, and is shipping a verification tool so anyone can check whether an image came from a model. The move is either a genuine step toward content provenance or an extremely sophisticated way of making it someone else's problem when the watermark gets stripped — possibly both simultaneously. It's worth noting that SynthID watermarks are invisible to the human eye, embedded in pixel-level statistical patterns, and have already been shown to survive moderate JPEG compression and cropping in some configurations. Security teams working on disinformation detection and incident response involving synthetic media should probably know which watermark standards their tools can and can't detect — because "it's watermarked" is only useful if someone checks.