cybr.cx Daily Digest — August 03, 2026

Critical Vulnerabilities

⚠️ Actively exploited — CVE-2026-20316 | Cisco Secure Firewall Management Center (FMC) | CVSS: N/A
Cisco's Secure Firewall Management Center (formerly Firepower Management Center) ships with a hard-coded password that allows an unauthenticated remote attacker to log in via a low-privileged account and access sensitive data. CISA added this to the KEV catalogue on July 29 with a remediation deadline that has already passed — if you haven't patched or isolated FMC instances, treat this as an active incident response situation. The hard-coded credential attack surface is trivially exploitable with no preconditions.

⚠️ Actively exploited — CVE-2025-68686 | Fortinet FortiOS | CVSS: N/A
This FortiOS flaw allows a remote unauthenticated attacker to bypass the patch Fortinet previously issued for its symbolic link persistence mechanism — meaning attackers with prior post-exploit footholds can re-establish access even on supposedly remediated devices. Crafted HTTP requests are sufficient to exploit this, and CISA's due date of August 10 is close. If your organisation patched earlier FortiOS symbolic-link issues and considered the matter closed, reconsider that posture immediately.

⚠️ Actively exploited — CVE-2026-16812 | Arista VeloCloud Orchestrator | CVSS: N/A
An OS command injection vulnerability in Arista's VeloCloud SD-WAN Orchestrator (on-premises) allows remote attackers to access privileged internal functionality, with full confidentiality, integrity, and availability impact on the orchestrator and connected SD-WAN infrastructure. CISA's remediation deadline has already passed (July 30). SD-WAN orchestrators are high-value targets — a compromised VCO gives an attacker visibility and control across the entire SD-WAN fabric.

CVE-2026-67356 | ArcadeDB < 26.7.3 | CVSS: 8.8
Schema-admin users in ArcadeDB can create JavaScript triggers that invoke getSecurity().createUser() without any permission checks, because the full LocalDatabase object is exposed to trigger contexts with HostAccess.ALL. Any user holding UPDATE_SCHEMA permission can escalate to a server-wide admin silently. Update to 26.7.3 and audit existing triggers.

CVE-2025-71399 | Better Auth < 1.4.5 | CVSS: 8.6
The underlying rou3 router normalises duplicate slashes, meaning //api/path and /api/path resolve identically — allowing attackers to bypass disabledPaths blocks and path-based rate limits simply by doubling a leading slash. Any application relying on Better Auth's path-based security controls for endpoint protection or brute-force mitigation is exposed until upgraded to 1.4.5.

CVE-2026-68581 | Vikunja 0.22.0–2.3.0 | CVSS: 8.1
Vikunja's API token management conflates user IDs and link-share IDs because both are independent numeric sequences resolved through the same generic interface. An attacker whose link-share JWT happens to carry a numeric ID matching a target user's ID gets treated as that user at the /api/v1/tokens endpoints — enabling token theft and account takeover. The fix requires proper principal-type validation; upgrade immediately.

CVE-2026-68580 | FreeRDP < 3.29.0 | CVSS: 7.5
A missing bounds check on the FramesPerPacket parameter received from RDP servers in FreeRDP's audio input channel (audin) causes an integer overflow leading to heap buffer overflow on ALSA backends, or denial of service on WinMM, sndio, and OpenSL ES. A malicious or compromised RDP server can trigger this against connecting clients. Update to 3.29.0, and note that this is a client-side risk triggered by server-supplied data.

CVE-2026-67357 & CVE-2026-68578 | ArcadeDB < 26.7.3 | CVSS: 7.5 each
Two more ArcadeDB MCP-transport flaws round out a bad patch cycle for this graph database. CVE-2026-67357 leaks the HA cluster token in cleartext through the get_server_settings MCP tool, which can then be used with spoofed headers to impersonate root. CVE-2026-68578 silently drops all permission checks for non-root MCP users, allowing arbitrary writes, schema mutations, and JavaScript execution. If you're running ArcadeDB with MCP enabled, 26.7.3 is not optional.

Headline News

Iranian Actors Linked to Coordinated Cyberattacks on U.S. Water Infrastructure

Michigan has disclosed that nine of its water utility systems were targeted in coordinated cyberattacks, joining Minnesota in a growing pattern of intrusions against U.S. water and wastewater infrastructure that intelligence officials attribute to Iranian threat actors. The attacks represent a deliberate focus on operational technology environments where disruption carries direct public safety implications, and come in the wake of prior FBI warnings to water sector operators to harden remote access and industrial control system interfaces. For practitioners, the pattern is consistent with Iranian groups probing critical infrastructure for both espionage value and coercive leverage rather than purely destructive intent — at least at this stage. OT security teams operating in water, energy, or municipal infrastructure should treat the FBI advisory as a live operational warning: review internet-exposed HMI and SCADA interfaces, enforce MFA on remote access, and ensure incident response plans account for both IT and OT recovery paths.

Word Document Worm Weaponises Microsoft Copilot to Self-Propagate

A researcher has demonstrated a worm that abuses Microsoft Copilot's document summarisation and generation capabilities by embedding instructions in hidden white text within Word documents. When Copilot processes the poisoned document, it follows the hidden instructions to silently alter numerical data in the output — changing figures, financial values, or statistics — and copies the malicious payload into the newly generated file, enabling propagation across any document the AI subsequently produces. The attack is a concrete example of indirect prompt injection moving from proof-of-concept curiosity to a plausible enterprise threat vector: the victim sees a clean document, Copilot sees a command surface. Practitioners should be aware that AI-assisted document workflows now represent an injection attack surface analogous to SQL injection in a different substrate — input sanitisation, output validation, and controlled Copilot deployment scopes all become relevant defensive controls.

Schrödinger's Feed

No quantum computing developments landed in the feeds today — the qubits appear to have decohered before reaching us. As a standing reminder: NIST's finalised post-quantum cryptography standards (ML-KEM, ML-DSA, SLH-DSA) are published and implementation timelines are not getting longer. Practitioners who haven't begun cryptographic inventory work — identifying where RSA and ECC are load-bearing in their infrastructure — are running out of comfortable runway. The threat isn't necessarily here today, but the migration effort almost certainly is.

/dev/random

A project called Kakehashi has appeared that attempts to run native macOS ARM binaries on Linux ARM in userspace — essentially a compatibility layer that translates macOS system calls and Mach-O binary formats well enough to execute real macOS executables on a kernel that has never seen an Apple. It's an impressive piece of plumbing, somewhere between Wine and a fever dream, and it raises the quietly interesting security question of what happens to macOS's security assumptions — code signing, entitlement checks, Gatekeeper — when the operating system enforcing them simply isn't there. We're not saying anyone should run untrusted macOS binaries through an experimental Linux shim, but we're also not not saying that.