August 21 Cluster: Four Critical Deadlines Across Microsoft, Apple, and Broadcom

Three vendors, four CVEs, one deadline — Friday, August 21. This cluster represents arguably the highest-density remediation sprint in recent KEV history and touches network infrastructure, virtualization, collaboration, and endpoint simultaneously.

CVE-2026-33824 — Microsoft IKE Service Extensions Double Free (RCE): A double-free memory corruption bug in the IKE Service Extensions stack enables remote code execution. IKE is foundational to IPsec VPN functionality on Windows systems. Internet-facing VPN concentrators and Windows servers with IKE enabled should be treated as unauthenticated RCE candidates until patched. Prioritize internet-exposed instances; consider temporarily disabling non-essential IKE endpoints if patching cannot be completed by Friday.

CVE-2026-55040 — Microsoft SharePoint Weak Authentication (Auth Bypass): Unauthenticated network attackers can bypass SharePoint's security controls entirely. For federal contractors operating SharePoint on-premises — still a substantial population under M365 hybrid deployments — this represents a direct data exfiltration risk. Cloud-hosted SharePoint tenants should confirm with Microsoft whether mitigations are applied at the platform level. On-premises deployments need patches applied immediately; if delayed, restrict SharePoint access to known IP ranges and enforce MFA at the network layer.

CVE-2026-59310 — Broadcom VMware vCenter Path Traversal (RCE): Network-accessible vCenter servers remain one of the most targeted assets in enterprise environments, and this path traversal allowing arbitrary code execution continues that pattern. Threat actors with even basic network access to the vCenter management plane can achieve full hypervisor-layer compromise — impacting every hosted VM simultaneously. Segment vCenter management interfaces from production networks if not already done, audit recent vCenter access logs for anomalous API calls, and apply Broadcom's patch before Friday. This is not optional maintenance.

CVE-2026-65400 — Apple macOS Improper Authentication (Screen Sharing Bypass): A network-adjacent attacker can authenticate to macOS Screen Sharing without valid credentials. In enterprise Mac fleets — common in engineering, design, and federal agency environments — this opens full graphical remote access to unpatched endpoints. Disable Screen Sharing on endpoints where it is not operationally required. For those that need it, enforce network-level firewall rules restricting VNC/Screen Sharing ports (5900/TCP) to authorized management subnets only, pending patch deployment.

AI/ML Infrastructure Under Active Threat: Developer Toolchains in the Crosshairs

The inclusion of MLflow and Ray in this KEV batch is a significant signal. These are not obscure edge tools — they are production-grade MLOps and distributed compute frameworks embedded across federal AI initiatives and commercial AI pipelines.

CVE-2026-64849 — MLflow SSRF (Deadline: September 2): This server-side request forgery in MLflow allows attackers to reach internal services and cloud metadata endpoints, receiving both response status and body content. In cloud-hosted ML environments, successful SSRF against the instance metadata service (IMDS) yields IAM credentials and can enable full account takeover. Immediately restrict MLflow server exposure to internal networks only, enforce IMDSv2 on AWS instances (blocking most SSRF-based metadata harvesting), and audit MLflow server logs for unexpected outbound requests to RFC-1918 or 169.254.169.254 addresses.

CVE-2025-62593 — Ray Code Injection / RCE (Deadline: August 21 — IMMINENT): Ray's code injection vulnerability is particularly alarming because exploitation paths run through Firefox and Safari — meaning developers browsing while running local Ray clusters are plausible targets. This blurs the line between developer workstation security and infrastructure security. Patch Ray immediately across all environments; do not treat development or staging deployments as lower priority. Developers running Ray locally should update browsers as a secondary control and avoid browsing untrusted sites while Ray dashboard ports are exposed.

Taken together, these two entries suggest adversaries are actively mapping and exploiting the AI development pipeline as a lateral movement pathway into broader cloud environments. MLOps security is no longer a future concern.

Operational Reminder: All eight entries carry BOD 26-04 obligations, including forensic triage requirements for confirmed exploitation. Agencies and federal contractors must document remediation status and report to CISA per applicable timelines.

Sources: CISA KEV Catalog · CISA BOD 26-04 · Cisco Security Advisory: ASA/FTD · Microsoft Security Update Guide · Broadcom VMware Security Advisories · Apple Security Releases · MLflow Security Advisories · Ray Project Security