August 20, 2026 is shaping up as a high-tempo day for defenders. CISA has pushed five new entries into its Known Exploited Vulnerabilities catalog in a single cycle, spanning AI/ML platforms, core Windows networking services, and collaboration software. Layered on top are a sophisticated ransomware double-extortion scam, a massive healthcare data breach, and a newly disclosed hardware-level side-channel attack against cloud serverless environments. The common thread: attackers are moving fast across every tier of the modern enterprise stack, and defenders need to match that pace.

The most urgent item is CVE-2026-64849, a Server-Side Request Forgery vulnerability in MLflow that CISA has confirmed is under active exploitation. MLflow is widely deployed in data science and machine learning pipelines, often with broad internal network access and minimal perimeter hardening because teams treat it as a developer tool rather than a production service. SSRF in this context is particularly dangerous: an attacker who can coerce the MLflow server into making internal HTTP requests can pivot to cloud metadata endpoints, internal APIs, secret stores, and adjacent services — all without direct network access. If your organization runs MLflow, treat this as a patch-now situation. In the interim, restrict MLflow's outbound network access using host-based firewall rules or network segmentation, disable unauthenticated access, and audit logs for unexpected outbound connections from the MLflow process.

CISA's second batch adds three more critical entries. CVE-2026-33824 is a Double Free vulnerability in the Microsoft Internet Key Exchange (IKE) Service Extensions. IKE is the handshake engine behind IPsec VPNs and is exposed on virtually every Windows server and endpoint that participates in site-to-site or remote-access VPN. A double-free condition at this layer can lead to memory corruption and, in the worst case, unauthenticated remote code execution against a service running with high privilege. CVE-2026-55040 affects Microsoft SharePoint and is described as a weakness vulnerability — a class of flaw that typically allows privilege escalation or authentication bypass within the SharePoint farm. SharePoint remains one of the most targeted Microsoft workloads because it aggregates sensitive documents and is frequently internet-facing. Both vulnerabilities should be prioritized for emergency patching ahead of the next scheduled maintenance window. Monitor IKE/ISAKMP traffic (UDP 500 and 4500) for anomalous negotiation patterns, and review SharePoint Unified Audit Logs for unexpected permission changes or bulk document access.

On the threat-actor creativity front, a suspected ransomware affiliate operating under the name Ransom Busters is contacting ransomware victims before attacks become public knowledge, impersonating a legitimate recovery firm and offering to provide decryption keys and data deletion for a fee. This is a double-extortion refinement: victims who pay receive nothing of value, and the payment itself may fund further attacks. The tactic exploits the chaos and urgency of the immediate post-incident window. Security leaders should brief incident response teams and legal counsel now, before an incident occurs, so that any unsolicited recovery offers are immediately escalated for vetting. Legitimate decryption assistance firms do not cold-contact victims, and any firm claiming to have obtained keys directly from a threat actor should be treated as suspect. Verify credentials independently and engage only through established IR retainer relationships.

The CareCloud breach, now confirmed to affect 3.7 million patients, is a stark reminder that healthcare IT vendors are high-value targets. Patient records contain a dense combination of PII, insurance data, and clinical history that commands premium prices on criminal markets and enables long-tail fraud. Organizations that rely on third-party health IT platforms should review their vendor risk management posture, confirm breach notification timelines in contracts, and ensure that data minimization principles limit what any single vendor can access or store.

Finally, researchers have demonstrated a remote Spectre-class side-channel attack against Cloudflare Workers that successfully leaked a JSON Web Token from a co-located Worker at up to 12 bits per second — a 360-fold improvement over prior techniques. While the bit rate sounds slow, a 256-bit JWT secret could theoretically be reconstructed given sufficient time and stable co-location. Cloudflare has been notified and is working on mitigations, but this finding has broader implications for any shared-tenant serverless or container platform. Defenders running sensitive workloads in multi-tenant cloud environments should evaluate whether JWTs and other short-lived secrets are the right trust mechanism for high-value service-to-service calls, consider rotating JWT signing keys more frequently, and monitor for anomalous same-origin traffic patterns that could indicate timing-based probing.

Defensive priorities for today:

- Patch CVE-2026-64849 in all MLflow deployments immediately; isolate MLflow from internal metadata services if patching is delayed
- Apply Microsoft patches for CVE-2026-33824 (IKE) and CVE-2026-55040 (SharePoint) on an emergency basis; monitor VPN gateway logs and SharePoint audit trails
- Brief IR and legal teams on the Ransom Busters scam; establish a verification checklist for any unsolicited recovery contact
- Audit third-party health IT vendor access and data retention as CareCloud breach details continue to emerge
- Review JWT lifecycle management and secret rotation cadence for workloads running on shared serverless platforms
- Ensure CISA KEV catalog entries are mapped to your asset inventory and tracked to closure within required timelines

This briefing is informational and does not replace official vendor advisories or CISA guidance — always consult primary sources before making patching and configuration decisions.