Welcome to the Cisco AI defense success capsule. In this capsule, we will review the main automation options available for Cisco AI defense. We will cover three options, the Cisco AI defense CLI, the Cisco AI defense Python SDK, and the.00:04Cisco AI defense Management API.00:20The goal is to help developers, security teams, dev secops teams, platform engineers, automation engineers, and partners understand what each option is, where it fits, and when to use it. By the end, you should have a practical decision framework for choosing the right automation approach for your AI defense use case.00:22Here is what we will cover. 1st, we will briefly discuss why automation matters for AI security adoption. Second, we will review the three automation options, the CLI, the Python SDK, and the management API. 3rd, we will compare the options and provide simple guidance on when to use each one.00:41This capsule is introductory and adoption focused. We will use examples where helpful, but we will keep the discussion short and practical. AI adoption often moves quickly. Developers maybe working with model files, model repositories, AI applications, agents, MCP servers, and multiple LLM provision.00:59If security checks only happen manually or only through a user interface, teams may struggle to keep up. Automation helps bring Cisco AI defense capabilities into the workflows that teams already use. This can include local development, CICD pipelines, Python applications, internal platforms, reporting systems, and custom integrations. E.g..01:19People, a developer might scan a model before sharing it. A devsecops team might add scanning to a build pipeline. A platform team might automate application onboarding. A security team might retrieve runtime events for reporting or investigation. The key benefit is that automation helps make AI security repeatable, scalable, and easier to adopt.01:39Cisco AI defense provides multiple automation options because different teams work in different ways. The CLI is a lightweight command line option. It is best when you wanna run model scans from a terminal, script or CICD pipeline. The Python SDK is a developer friendly library for Python applications and workflows.02:01It is best when you want to embed AI defense capabilities directly into application code, AI agents or Python automation logic. The management API is the REST API for programmatic administration and platform automation. It is best when you need direct API access for application onboarding, connection management, policy operations, validation jobs.02:19Event reporting.02:40Scan results and broader integrations. These options are complimentary. The right choice depends on the workflow, the user, and the level of integration needed. The 1st option is the Cisco AI defense CLI. The CLI provides a simple, developer friendly way to scan AI models for security risks from the command line.02:41It is especially useful for teams that want fast, on demand model scanning without using a UI driven workflow. Typical users include developers, devsecops engineers, platform engineers, security engineers, and automation specialists. The CLI supports scanning locally available model artifacts and scanning models hosted in HuggingFace reposit.03:01Repositories. It can also retrieve scan results in metadata. Because the CLI returns JSON output by default, teams can parse results programmatically and use them in scripts or CICD workflows. A simple way to think about the CLI is this.03:21Choose the CLI when you want quick repeatable model scanning from a terminal script or pipeline. Here are a few examples of how the CLI can be used. A developer can scan a local model file by running aidcly scan run-F followed by the model path. A security or platform team can scan a hugging.03:35This repository by using the repo URL option. This allows supported model files in the repository to be scanned without 1st downloading them locally. After scans are submitted, users can list recent scans and retrieve detailed scan information by scan ID. The CLI requires an AI defense management API key.03:55And the correct regional base URL for the environment. It is also important to note that CLI based scanning is cloud based. The CLI is the command line interface while the analysis runs in the Cisco AI defense cloud service. Local only scanning is not currently supported. The CLI binary is available for Linux X 80 six_sixty four systems.04:132nd option is the Cisco AI defense Python SDK. The SDK is best for developers and platform teams that want to integrate AI defense capabilities directly into Python applications or workflows. Compared with the CLI, the SDK provides a broader developer integration service.04:34It can support runtime protection, inspection workflows, model scanning, MCP scanning, validation, AI bomb related workflows, and management operations. One of the most important SDK capabilities is agent runtime protection through agent sec. This allows developers to add AI defense protection to supported LLM and MCP client.04:52At workflows with minimal application changes. The SDK is also helpful when developers want Python native objects, methods, error handling, and configuration patterns.05:12Instead of building directly against REST endpoints, a simple way to think about the SDK is this. Choose the SDK when AI defense needs to become part of your Python application or Python automation logic. Here are a few practical examples for the Python SDK. The package can be installed with pip install Cisco A defense.05:22SDK.05:42The primary Python import name is A defense. For runtime protection, a common pattern is to import agent sec from a defense runtime, and call agent set.protect with an agent sec YAML configuration file. After protection is enabled, the application can continue using supported LLM clients and MCP clients while AI defense inspection are.05:43Gateway routing is applied depending on the selected integration mode. The SDK supports API mode and gateway mode. In API mode, the SDK sends side channel inspection requests to Cisco AI defense while the application calls the provider directly. In gateway mode, traffic is routed through the Cisco AI defense gateway.06:03The SDK also includes clients for direct chat inspection, model scanning, management operations, AI validation, and MCP scanning. For production use, Teams should review the SDK documentation, manage secrets securely, and explicitly choose the desired integration mode and failure behavior.06:22The 3rd option is the Cisco AI defense Management API. The management API is a RESTful API that allows developers and administrators to interact programmatically with AI defense platform services. It is useful when teams need direct REST based automation, especially for systems that are not Python specific.06:41The API can be used to manage applications, connections, policies, runtime events, AI validation jobs, validation results, scan status, and other platform resources. It can also support broader integration patterns such as internal portals, reporting pipelines, back end services, and operational dashboards, the open.06:59API specification also includes MCP registry and resource connections endpoints marked as beta. These can support MCP related workflows where applicable.07:19A simple way to think about the management API is this. Choose the management API when you need direct language independent rest automation for AI defense platform operations. Here are a few simple management API examples. To list applications, a client can call get applications under the AI defense API based.07:28Graph.07:48To start an AI validation job, a client can call post AI validation/start to retrieve scan status and findings, a client can call get scan/scan ID. Management API calls require an AI defense management API key. The key is generated in the AI defense UI under administration and API keys. Each API.07:48API request includes the key in the X Cisco AI defense tenant API key header. The management API key is not interchangeable with an AI defense inspection API key.08:08This distinction is important. Management API keys are used for management and automation workflows. Inspection API keys are used for runtime inspection workflows. Regional base URLs are also important. E.g., the North America base URL is HTTPS colon/slash aPI.us.security.cisco.08:18Go.com. APAC uses HTTPS colon/slash API.apj.security.cisco.com. Europe uses HTTPS colon/slash API.EU.security.cisco.com. As a best practice, do not hard code API keys in production code.08:38Use environment variables, vaults or approved secret management systems. So how do you choose? Use the CLI when the workflow is command line driven. This is ideal for quick model scans, scripts, and CICD pipeline gating.08:57Use the Python SDK when the workflow is application driven and Python based. This is ideal when developers want to integrate AI defense into AI applications, agent workflows, model scanning logic, validation workflows or custom Python automation. Use the management API when the workflow is platform driven or integration driven.09:14This is ideal for rest based automation, backend services, internal tools, dashboards, reporting workflows, and systems that need direct access to AI defense platform resources. There is also overlap. E.g., model scanning can be done through the CLI, SDK or API.09:34Depending on the workflow, the CLI maybe easiest for pipeline scripts. The SDK maybe better for Python automation. The management API maybe better for centralized platform integrations. The key is to start with a workflow, then choose the automation option that fits that workflow best.09:51Let us summarize the Cisco AI defense CLI is a lightweight option for command line model scanning. It is a strong fit for developers, devsecops teams, and security teams that want fast repeatable scans and local workflows or CICD pipelines.10:09The Cisco AI defense Python SDK is best when AI defense needs to be integrated directly into Python applications, AI agents or automation workflows. It provides Python native integration for runtime protection, inspection, scanning, validation, and management use cases. The Cisco AI defense management API is best.10:24For direct rest based platform automation.10:44It is useful for application onboarding, connection management, policy operations, event reporting, validation workflows, scan reporting, and broader integrations. These options are complimentary. The best choice depends on who is automating, what workflow they are automating, and how deeply AI defense needs to be integrated. Thank you for watching the Cisco AI.10:46Defense success capsule.11:06
