IMPERIAL CYBERX← AI Security
AI MODEL / GEMMA 4

AI Gemma 4.
26B A4B.

Google Gemma 4 26B A4B is an open model available through Cloudflare Workers AI. Imperial CyberX uses the model identifier @cf/google/gemma-4-26b-a4b-it in its AI CyberCafe backend.

MODEL REFERENCE

What is
Gemma 4?

Gemma 4 26B A4B is a Google model with a Mixture-of-Experts architecture. Cloudflare documents 26 billion total parameters with approximately 4 billion active per forward pass, providing a model architecture designed around efficient inference.

Through Workers AI, the model can be used for text generation and structured application workflows, with documented support for long context, vision, reasoning, batch processing, and function calling.

CAPABILITIES

Built for
modern AI workflows.

01

Long context

A documented 256,000-token context window supports long conversations, large documents, tool definitions, and extended workflows.

02

Vision

Gemma 4 26B A4B supports vision workloads including document, PDF, image, screen, chart, OCR, and handwriting understanding.

03

Reasoning

The model supports reasoning workflows and built-in thinking capabilities for tasks that require additional inference depth.

04

Function calling

Native function calling supports structured interactions between the model and controlled application tools.

SECURITY

Model capability
is only part of
the system.

01

Prompt boundaries

Define trusted instructions, untrusted inputs, and application-level controls around model requests.

02

Data handling

Review sensitive data paths through prompts, context, retrieval systems, logs, and outputs.

03

Tool permissions

Constrain model-initiated tool use with explicit authorization, validation, and execution boundaries.

04

Application controls

Treat the model as one component of a larger application security architecture rather than a standalone control.

IMPERIAL CYBERX / AI

Explore AI security
around the model.

Model selection is one layer of an AI system. Security also depends on prompts, data flows, permissions, integrations, application controls, and the surrounding workflow.

AI Security →AI Assessment →