Plugin4Shell flaw exposes Claude Code, Codex, Copilot and Gemini CLI to plugin verification bypass

Plugin4Shell exposes a shared weakness in AI coding agent plugin verification

A vulnerability dubbed Plugin4Shell was disclosed in September 2026, and it targets the plugin verification path shared by four widely deployed AI coding assistants. According to the disclosure, Claude Code, OpenAI Codex, GitHub Copilot, and Gemini CLI are all affected, meaning that a single class of flaw now spans the agent tools many developers already have running in their terminals, editors, and CI runners. Because plugin installation is the route through which these agents extend their tool-use surface, a verification bypass has direct consequences for any developer who installs third-party plugins or allows them to be installed automatically by their team’s onboarding scripts.

Practitioners who maintain internal plugin registries or pin specific plugin versions should treat this as a high-priority item to verify once vendor patches land, and teams running agent workloads in CI should review their plugin allowlists while the patch cycle is still in motion.

Centralpoint DXP 8.11.109 advances data transfer capabilities

In a separate vendor announcement, Oxcyon released Centralpoint Digital Experience Platform version 8.11.109, with the company highlighting improvements to data transfer capabilities. The announcement was distributed as a press release from Bath, OH in July 2025, and it targets the digital experience platform category where enterprise teams assemble portals, intranets, and content hubs. For practitioners integrating Centralpoint with upstream and downstream systems, the 8.11.109 release is the version string to pin when reproducing reported behaviour or filing support tickets, since the data-transfer changes are the headline change rather than a routine maintenance bump.

Intel refreshes Volume 4 of the 64 and IA-32 software developer manual

Intel has updated its 64 and IA-32 Architectures Software Developer’s Manual Volume 4, the volume dedicated to model-specific registers. The manual sits in Intel’s Development Tools area alongside the Software Catalog and Download Center, and it is the canonical reference for the MSR encodings that operating-system kernels, hypervisors, firmware, and performance tooling depend on. Developers targeting low-level x86 code paths, including those building or maintaining AI accelerator host stacks, profilers, and bare-metal firmware, should compare this refreshed edition against any in-house MSR tables, since MSR definitions shift across microarchitectures and silent drift against an older printed copy can produce subtle miscompilations.

What practitioners should verify after the Plugin4Shell disclosure

The most concrete next step for working developers is to audit which of the four named agents are in active use across their environments and to inventory the plugins installed in each, since Plugin4Shell targets the verification step rather than the agent runtime itself. Where teams run agent workloads headlessly, pinning plugin versions and disabling auto-update of plugins is a defensible short-term posture until vendor patches are confirmed. Teams that ship their own agent plugin should review how their plugin payload is signed, hashed, and fetched, because a verification bypass upstream means downstream consumers cannot rely on the agent’s own signature check alone.

Practitioners who want a quick way to normalise plugin metadata or check hashes across registries may find a MIME Type Lookup useful when triaging plugin artefacts, and a URL Extractor can help when scraping plugin marketplaces to rebuild an allowlist from scratch. For organisations documenting the incident, a Merge Excel Files step can consolidate plugin inventories from multiple teams into a single review sheet ahead of the patch cycle.