cybr.cx | August 22, 2026

Critical Vulnerabilities

⚠️ Actively exploited — CVE-2026-73570 | Zimbra Collaboration Suite | CVSS: N/A
An OS command injection flaw in Zimbra ZCS allows unauthenticated attackers to send specially crafted SMTP requests and execute arbitrary operating system commands as the Zimbra user. CISA added this to KEV on August 21 with a remediation deadline of August 24 — that's 48 hours. If you're running ZCS on-premises, this is your top priority today.

⚠️ Actively exploited — CVE-2026-72529 & CVE-2026-72530 | TrueConf Server | CVSS: N/A
Two chained vulnerabilities in TrueConf Server are being actively exploited via port 4307/TCP. CVE-2026-72529 allows unauthenticated remote attackers to execute arbitrary scripts (missing authentication on a critical function); CVE-2026-72530 allows sandbox escape and arbitrary code execution on the host. Together, they represent a full unauthenticated RCE path. CVE-2026-72529's remediation deadline has already passed (August 23) — patch or block port 4307 externally now.

⚠️ Actively exploited — CVE-2026-59310 | VMware vCenter | Broadcom | CVSS: N/A
A path traversal vulnerability in vCenter allows any attacker with network access to the management interface to execute arbitrary code. With vCenter typically sitting at the heart of virtualisation infrastructure, successful exploitation can mean full datacenter compromise. Broadcom's patching cadence here has been watched closely — verify you're on the latest build.

⚠️ Actively exploited — CVE-2026-55040 | Microsoft SharePoint | CVSS: N/A
A weak authentication vulnerability in SharePoint allows unauthenticated network attackers to bypass security controls. SharePoint's prevalence in enterprise environments and its exposure through hybrid deployments makes this a high-value target. The CISA remediation deadline was August 21 — you're already overdue.

⚠️ Actively exploited — CVE-2026-65400 | Apple macOS | CVSS: N/A
An improper authentication vulnerability in macOS Screen Sharing allows network-adjacent attackers to authenticate without valid credentials. Any Mac with Screen Sharing enabled and network exposure is potentially accessible to unauthenticated attackers. Update macOS immediately and audit which machines have Screen Sharing enabled.

⚠️ Actively exploited — CVE-2026-33824 | Microsoft IKE Service Extensions | CVSS: N/A
A double-free vulnerability in Microsoft's Internet Key Exchange service enables remote code execution. IKE is foundational to IPsec VPN infrastructure — exploitation here could undermine encrypted tunnel integrity or enable lateral movement through VPN concentrators.

⚠️ Actively exploited — CVE-2026-64849 | MLflow | CVSS: N/A
An SSRF vulnerability in MLflow allows attackers to reach internal services and cloud metadata endpoints, retrieving response status and body. MLflow is widely deployed in ML pipelines, including environments with AWS/GCP/Azure IMDS exposure — this is a direct path to credential theft from cloud metadata services.

⚠️ Actively exploited — CVE-2025-62593 | Ray (Ray-Project) | CVSS: N/A
A code injection vulnerability in the Ray distributed computing framework enables remote code execution, exploitable via Firefox and Safari. Developer environments running Ray are in scope, making this a supply chain risk vector for organisations building ML infrastructure.

CVE-2026-77775 | Headroom LLM Proxy | CVSS: 8.6 (HIGH)
Headroom's LLM proxy trusts the client-supplied x-headroom-base-url header to select the upstream destination, performing only a superficial scheme and hostname check. This is a server-side request forgery vulnerability — attackers can redirect proxied LLM traffic to internal infrastructure. Environments exposing Headroom publicly should treat this as critical regardless of the CVSS score.

CVE-2026-72860 | 9router | CVSS: 8.5 (HIGH)
The /api/provider-nodes/validate route in 9router issues server-side HTTP requests to attacker-controlled URLs. The SSRF guard only compares hostname strings, skips DNS resolution, ignores redirect chains, and has a broken IPv4-mapped IPv6 bypass branch — making it essentially decorative. Full SSRF to internal services is achievable.

CVE-2026-39909 | llama.cpp (before b8585) | CVSS: 8.1 (HIGH)
A use-after-free in llama.cpp's RPC server GRAPH_RECOMPUTE handler allows unauthenticated remote attackers to achieve arbitrary read/write by storing a computation graph, freeing its buffers, then reclaiming that memory with attacker-controlled content. Anyone running llama.cpp with the RPC server exposed — even on internal networks — should update to b8585 immediately.

CVE-2026-17006, CVE-2026-17060, CVE-2026-19437 | IBM AIX 7.2/7.3, PowerVM VIOS 4.1 | CVSS: 8.1–8.3 (HIGH)
A cluster of remote vulnerabilities in IBM AIX and PowerVM VIOS: heap buffer overflow enabling RCE (CVE-2026-17006, 8.3), kernel heap over-read causing information disclosure and DoS (CVE-2026-17060, 8.1), and a second buffer overflow with RCE potential (CVE-2026-19437, 8.1). AIX environments tend to be high-value targets in financial and government sectors — apply IBM's fixes and verify network-level access controls on affected systems.

Headline News

Cognizant Data Breach Exposes Customer Personal Information
Cognizant, the multinational IT and professional services firm headquartered in Teaneck, NJ, suffered a data breach on or around April 21 that may have exposed customers' personal information. The company is now offering protection services to affected individuals — a standard post-breach response that nonetheless signals the breach has been assessed as high enough severity to warrant notification obligations. For practitioners, Cognizant's role as a major managed services and outsourcing provider means downstream exposure is worth assessing: if your organisation shares data with Cognizant or uses its platforms, review what was in scope. Incidents at large MSPs are rarely contained cleanly — supply chain visibility remains a persistent blind spot across the industry.

Texas Student Exposes Rogue AI Hacking Attempt
Roughly two weeks ago, a Texas student identified and reported what appears to be an unauthorised AI-assisted hacking attempt — a case now drawing attention for what it signals about AI being turned against educational and institutional infrastructure. Details remain sparse, but the incident underscores an emerging threat pattern: AI tooling lowering the barrier for attack attempts by less-skilled actors who can leverage automated capabilities to probe or exploit systems they wouldn't previously have had the skills to target. For defenders, the episode is a reminder that AI-augmented attacks are no longer theoretical — threat models should account for automated reconnaissance, exploit generation, and social engineering at scale. Student-facing systems and academic networks, frequently under-resourced from a security standpoint, are increasingly attractive targets.

Schrödinger's Feed

Canadian firm Crypto4A Technologies has achieved what it's calling a world first: FIPS 140-3 Level 3 validation for a quantum-safe hardware security module, with the QASM™ cryptographic module supporting the full suite of NIST-standardised post-quantum cryptography algorithms. FIPS 140-3 Level 3 is a meaningful bar — it requires physical tamper resistance and identity-based authentication, not just algorithmic compliance. This means organisations can now procure a validated, production-grade HSM that natively supports ML-KEM, ML-DSA, and SLH-DSA without bolting PQC onto legacy hardware. Practitioners building crypto agility roadmaps or facing compliance timelines around quantum-readiness now have a certified hardware option to evaluate — the "we're waiting for validated implementations" argument is getting harder to sustain.

/dev/random

A study tracked students before and after access to AI homework assistance and found a consistent pattern: homework scores went up, exam scores went down. The working theory is that AI tools are optimising for output — completed assignments — rather than for the learning process that makes exams passable without a chatbot in the room. It's a tidy demonstration of Goodhart's Law applied to education: when the measure becomes the target, it ceases to be a good measure. Security teams who've watched AI-assisted code pass review only to fail spectacularly in production will find the dynamic grimly familiar.