Citrix patches two actively exploited NetScaler RCE zero-days as encoding-layer flaws surface across Cloudflare and Oracle stacks

Citrix NetScaler zero-days hit default configurations

Two previously unpatched Citrix NetScaler ADC and NetScaler Gateway vulnerabilities allow remote code execution and have been exploited in the wild, Citrix confirmed on September 27. The vendor issued patches for both alongside fixes for six other flaws. One of the two critical bugs affects every deployment running an affected version, including those in the default configuration, which compresses the response window for teams that have not segmented or hardened beyond defaults. The disclosure arrived one day after a security firm publicized its findings.

URL-encoding bypass revives Oracle PeopleSoft attacks

ShinyHunters is using a URL-encoding trick to slip malicious payloads past web application firewall rules intended to mitigate CVE-2026-35273, letting the group resume widespread exploitation of vulnerable Oracle PeopleSoft servers. The technique points to a familiar encoding-layer weakness: defenders who signature on decoded strings are now being routed around by adversaries who weaponize the encoding scheme itself rather than the payload underneath. Teams running WAFs in front of PeopleSoft should treat encoder normalization as a first-class control and verify that their rule set inspects percent-encoded forms before forwarding upstream.

Cloudflare patches a cross-tenant data leak in Workers Containers

Cloudflare fixed a vulnerability in Containers and Sandboxes that let a customer with a Workers Paid account recover residual data from other customers' containers sharing the same physical host. For teams that isolate tenants through container boundaries, the incident is a reminder that residual data hygiene and host-level cleanup are part of the encoding-and-serialization contract, not a separate concern. Affected accounts should confirm which container images and cached objects were live during the exposure window.

A four-stage MaaS stealer abuses an AMD driver to silence defenses

The Lunex stealer, part of the Psychedelic Stealer malware-as-a-service platform, is being distributed through compromised Ukrainian websites using ClickFix-style fake Cloudflare CAPTCHA pages. A four-stage attack chain ultimately leverages an AMD driver to disable security monitoring before harvesting browser credentials from Ukrainian-speaking users. The encoding angle here is the social-engineering shell: the bogus verification page normalizes a multi-stage download on the victim side, with telemetry obscured by signed-driver abuse.

Anthropic consolidates AI tooling into a single Claude Marketplace

Anthropic launched a Claude Marketplace that aggregates plugins, connectors, agents, and related AI tools in one place, exceeding 2,000 entries. For practitioners integrating encoded or hashed data into model pipelines, the marketplace creates a single distribution point to audit for cryptographic handling, signing, and transport security before adopting any connector.

Practitioner follow-ups to check

- Apply the September 27 Citrix NetScaler fixes immediately and audit for indicators of pre-patch exploitation, particularly on appliances in default configurations. - Re-test WAF rule coverage against percent-encoded variants of payloads targeting CVE-2026-35273, and confirm upstream decoders normalize before signature matching. - Confirm with Cloudflare which Workers Containers or Sandboxes workloads ran on the affected hosts, and rotate any keys, tokens, or cached payloads that could have leaked.

- Segment Ukrainian-language web traffic policy the same way as credential-theft traffic, and watch for ClickFix-style Cloudflare CAPTCHA impersonations that escalate through signed-driver abuse. - Inventory any connector or agent pulled from the new Claude Marketplace and review its transport, signing, and identity model before granting production data access.