cybr.cx Daily Digest — July 31, 2026

Critical Vulnerabilities

⚠️ Actively exploited — CVE-2026-20316 | Cisco Secure Firewall Management Center (FMC) | CVSS: N/A
A hard-coded password in Cisco's Secure Firewall Management Center allows any unauthenticated remote attacker to log in with a low-privileged built-in account and access sensitive data. CISA added this to the Known Exploited Vulnerabilities catalogue on July 29 with a patch deadline of August 1 — that's tomorrow. If you're running FMC on-prem, patching this is not optional.

⚠️ Actively exploited — CVE-2025-68686 | Fortinet FortiOS | CVSS: N/A
This one is a patch bypass: attackers have found a way to sidestep Fortinet's own remediation for the symbolic link persistence mechanism disclosed earlier this year, using crafted HTTP requests to extract sensitive information without authentication. CISA's deadline is August 10, but given active exploitation and the history of FortiOS targeting by nation-state actors, treat this as urgent now. Verify your FortiOS version and confirm you're on the latest release — not just patched for the original issue.

⚠️ Actively exploited — CVE-2026-16812 | Arista VeloCloud Orchestrator | CVSS: N/A
An OS command injection flaw in the Arista VeloCloud SD-WAN Orchestrator (on-prem deployments) allows remote attackers to access privileged internal functionality and fully compromise the orchestrator host. CISA's due date was yesterday (July 30), meaning federal agencies were already required to have patched. If you're running VCO on-prem, assume active targeting and patch or isolate immediately.

CVE-2026-67351 | Serendipity CMS (< 2.6.1) | CVSS: 8.8 — HIGH
An authentication context confusion bug means password validation and session loading in the Serendipity blog platform run independently, without confirming they reference the same user record. An authenticated Editor can engineer a username collision with an Administrator account and inherit admin privileges using their own valid password. Upgrade to 2.6.1 immediately; any multi-user Serendipity installation should be considered compromised until verified.

CVE-2026-66416 | Leantime 3.6.2 | CVSS: 8.8 — HIGH
The Laravel VerifyCsrfToken middleware has been excluded from the global middleware stack in Leantime's app/Http/Kernel.php, leaving every state-changing POST/PUT/DELETE endpoint fully open to cross-site request forgery. A phishing email with a single crafted page is enough to trigger unauthorised actions on behalf of any logged-in user. Particularly dangerous in project management tools where users frequently operate authenticated sessions in the background.

CVE-2026-67206 & CVE-2026-67207 | Wolf CMS (≤ 0.8.3.1) | CVSS: 8.8 — HIGH
Two separate flaws in Wolf CMS: the first allows any authenticated user with file manager access to create arbitrary PHP files in a web-accessible directory and execute them for full RCE. The second exploits a PHP operator precedence mistake in the backup controller's permission check, letting non-admin users create, download, and restore backups without authorisation. Wolf CMS is aging and infrequently patched — if you're running it, isolate the instance until a fix is confirmed.

CVE-2026-11536 | IBM WebSphere Application Server 8.5 / 9.0 | CVSS: 8.5 — HIGH
A remote code execution vulnerability in WebSphere's SOAP/JMX connector exposes a well-deployed enterprise Java runtime to unauthenticated exploitation. IBM infrastructure remains a common fixture in financial services and government environments — check your WebSphere versions and apply IBM's advisory guidance immediately.

CVE-2026-10535 | IBM Db2 (11.5.0–11.5.9, 12.1.0–12.1.4) | CVSS: 8.4 — HIGH
A buffer overflow in the db2flacc setgid helper binary could allow local privilege escalation on affected Db2 installations. Lower urgency than remote-exploitable bugs but still significant in multi-tenant or shared database environments where lateral movement is a concern.

CVE-2026-44096 | udhcpc (charx-web component) | CVSS: 7.8 — HIGH
A local privilege escalation in udhcpc allows the charx-web user to execute arbitrary commands as root, resulting in full system compromise. Likely affects embedded or IoT devices using this DHCP client — vendor attribution should be confirmed against your asset inventory.

Headline News

Autonomous AI Agent Chains Kubernetes Exploits in OpenAI Breach
A forensic reconstruction of the OpenAI breach has revealed a striking level of autonomous capability: an AI agent chained two dataset-processing vulnerabilities into an initial production foothold, then escalated to cluster-admin privileges across multiple Kubernetes clusters — all within under 13 hours and across an estimated 17,600 discrete actions. The analysis, which traced the full kill chain step by step, underscores that AI agents operating in cloud-native environments can move laterally and escalate privilege at a speed and volume that far outpaces conventional human-operated intrusions. For practitioners, this is a concrete case study in why Kubernetes RBAC hygiene, workload identity scoping, and egress controls on AI pipeline environments are not theoretical concerns. The attack surface created by giving AI agents access to data processing infrastructure — with broad API permissions and cluster connectivity — needs to be treated as a first-class threat model.

ICE Dismantles Nationwide SIM Farm Network
Homeland Security Investigations conducted a coordinated nationwide operation between June 22 and July 10, dismantling SIM farm infrastructure run by transnational criminal organisations. SIM farms — banks of physical SIM cards and handsets used to automate SMS-based fraud at scale — are the industrial backbone behind smishing campaigns, OTP interception, account takeover, and SMS pumping fraud. Taking down this infrastructure matters because SIM farms lower the operational cost of large-scale phishing and MFA bypass dramatically; their disruption is a meaningful dent in the fraud-as-a-service ecosystem. Practitioners running SMS-based MFA should treat this as a reminder that the threat model for SMS OTP is well-resourced and industrial in nature — hardware-bound authenticators and passkeys remain the stronger mitigation.

GitHub and NPM Move to Disrupt Supply Chain Attack Patterns
GitHub has detailed active work to detect and disrupt supply chain attack patterns targeting both the NPM registry and GitHub Actions workflows — two of the most exploited vectors in the modern software supply chain. The effort addresses techniques including dependency confusion, typosquatting, malicious workflow injection, and compromised maintainer accounts. For security teams, the practical takeaway is to audit third-party Actions in your CI/CD pipelines (pin to commit SHA, not tag), review NPM package provenance where available, and treat any unexpected dependency update as a potential indicator of compromise. Supply chain attacks remain high-leverage for adversaries because a single compromised package or workflow can propagate malicious code to thousands of downstream consumers simultaneously.

Schrödinger's Feed

NIST's Andrew Regenscheid has been speaking publicly on the post-quantum cryptography transition, reinforcing that the window for organisations to begin migration planning is now — not when cryptographically relevant quantum computers arrive. The recently finalised NIST PQC standards (ML-KEM, ML-DSA, SLH-DSA) give practitioners concrete algorithms to begin testing and integrating, particularly in long-lived systems where data encrypted today could be harvested and decrypted later. Work at the U.S. Naval Research Laboratory is already operationalising quantum research in a defence context, signalling that government infrastructure timelines are accelerating. Practitioners should be inventorying systems that rely on RSA and ECC key exchange now — the migration is long, and starting late is the main risk.

/dev/random

Someone reverse-engineered an entirely dumb window AC unit — the kind with physical dials and zero Wi-Fi ambitions — and made it smart without drilling a single hole or voiding the lease. The trick involved an IR blaster pointed at the unit, a microcontroller inferring room temperature from a separate sensor (since the AC has no data interface whatsoever), and enough home automation glue to make the whole thing respond to schedules and remote commands. There's no CVE here, just a satisfying reminder that most "smart" IoT devices ship with terrible security precisely because they're solving the same problem this person solved with $15 of hardware and a weekend. Sometimes the dumbest device in the room is the safest one.