cybr.cx Daily Digest — August 13, 2026

Critical Vulnerabilities

⚠️ Actively exploited — CVE-2026-20349 | Cisco ASA / FTD | No CVSS in NVD feed
A heap inspection flaw in Cisco's Secure Firewall ASA and FTD products allows an unauthenticated remote attacker to crash the device, causing a denial-of-service. CISA added this to the KEV catalogue on August 11 with a remediation deadline of August 14 — tomorrow. If you're running perimeter Cisco firewall kit and haven't patched, you're already late.

⚠️ Actively exploited — CVE-2026-68820 | Microsoft Windows WinSock Driver | No CVSS in NVD feed
A use-after-free in the Windows Ancillary Function Driver for WinSock enables local privilege escalation for any authenticated attacker. This is the zero-day closed in August's Patch Tuesday drop — Microsoft shipped fixes for 398 vulnerabilities this cycle, and this one was already being weaponised before the patch landed. Post-exploitation chains involving this bug are the obvious concern; patch endpoints now, prioritise internet-facing Windows systems.

⚠️ Actively exploited — CVE-2026-72898 | Metabase | No CVSS in NVD feed
Unauthenticated SQL injection in Metabase grants a remote attacker full administrator access to the instance — from there, configuration changes, credential theft, and lateral movement are all on the table. CISA's deadline was August 14. If Metabase is exposed to the internet in your environment, treat this as a P0.

⚠️ Actively exploited — CVE-2026-8037 | Progress LoadMaster | No CVSS in NVD feed
Command injection via unsanitised input across multiple endpoints in Progress LoadMaster allows unauthenticated RCE on the appliance. CISA's remediation deadline already passed (August 10) — if you haven't acted, assume compromise and investigate accordingly.

⚠️ Actively exploited — CVE-2026-63077 | JetBrains TeamCity | No CVSS in NVD feed
Deserialization of untrusted data in TeamCity's agent polling protocol enables unauthenticated remote code execution. Build infrastructure is high-value for supply chain attackers; this vector has been abused before and is being abused again. Patch or isolate immediately.

CVE-2026-57858 | Cal.com Cal.diy (v2.1.1–6.2.0) | CVSS 8.9
Stored XSS in the BookingPageTagManager component allows any authenticated event owner to inject a malicious analytics tracking ID that executes arbitrary JavaScript in every visitor's browser. The payload escapes the inline script string literal — a classic but effective technique. Impact is broad given public-facing booking pages; update to a patched release and audit existing event configurations for suspicious tracking IDs.

CVE-2026-18099 | IBM i 7.3–7.6 | CVSS 8.9
Improper neutralisation of user-controlled input allows a remote authenticated attacker to execute arbitrary script code on IBM i systems. Multiple IBM i CVEs landed today (see also CVE-2026-16856, CVE-2026-17110, CVE-2026-18713, CVE-2026-17082, CVE-2026-17417) covering privilege escalation, OS command injection, and shell metacharacter abuse across versions 7.3 through 7.6. IBM i environments should treat this batch as a coordinated patching event — the attack surface across these issues is substantial.

CVE-2026-13361 | IBM Informix (oninit) | CVSS 8.8
An unchecked length field in the sq_sgkprepare SQL interface handler enables remote code execution on IBM Informix database servers. RCE via a database listener is a serious exposure; ensure Informix instances are not directly reachable from untrusted networks and apply vendor patches.

Headline News

Mass Vulnerability Scanning Camouflaged as AI Crawlers

An ongoing campaign of large-scale vulnerability scanning has been observed spoofing well-known AI bot user-agents — including ClaudeBot — to blend scanning traffic into the noise of legitimate AI web crawlers. The technique is deliberate: many organisations have begun allowlisting or deprioritising AI crawler traffic in WAF rules and log analysis, creating a blind spot that attackers are now actively exploiting. The scans appear broadly opportunistic, probing for known web application vulnerabilities across a wide target range rather than focusing on specific sectors. For defenders, the practical takeaway is to stop treating AI bot user-agents as inherently benign in security tooling — user-agent strings are trivially spoofed and should carry no implicit trust. Review WAF and SIEM rules that may be suppressing alerts for recognised crawler signatures.

Fake CCleaner Campaign Delivers GhostDesk Chrome Spyware

A convincing lookalike CCleaner website is serving a multi-stage malware dropper that ultimately installs a malicious Chrome extension dubbed GhostDesk, designed to conduct in-browser surveillance of victims. The campaign is notable for both the quality of the social engineering — the fake site closely mirrors the legitimate CCleaner brand — and the choice of a browser extension as the final payload, which persists across sessions and operates within the trusted browser context where credentials, session tokens, and sensitive data flow freely. GhostDesk's spyware capabilities allow exfiltration of browsing activity and potentially stored credentials, making it a significant threat to individuals and corporate environments alike. The campaign underscores the continued risk of typosquatting and brand-impersonation for popular utility software — a category users tend to download reactively when a machine feels slow. Practitioners should consider browser extension management policy and allowlisting as a baseline control, particularly in environments handling sensitive data.

Zoom Screen-Share Flaw Enabled Full Device Takeover — Discovered by AI

A vulnerability in Zoom's screen-sharing functionality could allow any participant on a call to seize control of another participant's device — a significant boundary violation in one of the world's most widely deployed conferencing platforms. The bug was discovered autonomously by an AI system conducting vulnerability research, representing one of the more concrete recent demonstrations of AI-driven offensive security capability operating without direct human direction at the discovery phase. The practical impact during the window the flaw was unpatched is sobering: any Zoom call where screen sharing was enabled could have been a vector for device compromise, affecting enterprise, government, and personal users equally. Zoom has issued a patch; users should ensure they are running the latest client version. The broader signal for practitioners is the accelerating pace at which AI tooling is finding exploitable flaws — a dynamic that will compress the time between vulnerability existence and active exploitation.

Schrödinger's Feed

Quantinuum's 98-qubit Helios trapped-ion processor is being deployed directly inside an Oracle Cloud Infrastructure data centre under a new multi-year partnership — bringing quantum hardware into closer proximity with classical cloud workloads than most enterprise users have previously encountered. Trapped-ion systems currently hold the edge in gate fidelity over superconducting competitors, which matters enormously for the error rates that determine whether cryptographically relevant computation remains a near-term or distant concern. The integration into a major commercial cloud platform also means organisations will soon have more accessible on-ramps to quantum experimentation — including testing post-quantum cryptography assumptions. Practitioners working on PQC migration timelines should factor accelerating hardware availability into their threat modelling; "harvest now, decrypt later" attacks are already underway, and the compute side of that equation is maturing faster than expected.

/dev/random

Tailscale's engineering team recently traced a database corruption issue in their infrastructure back to a bug in SQLite's WAL (Write-Ahead Logging) reset logic — a bug that has been sitting quietly in SQLite since 2008. The flaw relates to how SQLite handles WAL file resets under specific conditions, a code path that apparently went undetected for 16 years across one of the most widely deployed database engines on the planet. SQLite runs in an estimated trillion-plus deployments, from smartphones to aircraft, which makes a latent corruption bug in its durability mechanism a genuinely unsettling discovery. Credit to Tailscale for digging all the way down to the root cause rather than papering over it — though the real credit might go to whoever wrote subtly wrong WAL logic in 2008 and kept us all in suspense for a decade and a half.