# December 2025

## **AI Assets: MCP Servers**

As part of the AI Assets, in addition to Models and AI Workflows, [MCP Servers](https://docs.probe.splx.ai/ai-asset-management/mcp-servers) have now been added, expanding the ability to scan and manage MCP server infrastructure. The MCP Servers page provides a control panel for monitoring and managing MCP servers from the connected repositories, enabling users to assess server statuses, prioritize issues by severity, and take necessary actions.

<figure><img src="https://1029475228-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fi12bk7lo75SODuwcRCQp%2Fuploads%2FjFi2j2IIrVRim2zNsDu6%2Fmcp-server.png?alt=media&#x26;token=fbf5aaa3-95b8-48bd-90e1-2427b71d2f98" alt=""><figcaption><p>Figure 1: MCP Server Details</p></figcaption></figure>

## **AWS Bedrock Guardrail Policy Generator**

The Policy Generator has been upgraded to now support [the creation of policies for AWS Bedrock Guardrails](https://docs.probe.splx.ai/ai-red-teaming/remediation/policy-generator/aws-bedrock-guardrails). This improvement allows users to create and manage policies specifically designed to enforce security, compliance, or operational guidelines on AWS Bedrock services with greater ease and efficiency.

## Proxy SDK Connection <a href="#red-teaming-target-new-connection-type-proxy-sdk" id="red-teaming-target-new-connection-type-proxy-sdk"></a>

[Proxy SDK](https://docs.probe.splx.ai/ai-red-teaming/probe/target/index/proxy-sdk) has been added as a new connection type under the API category for configuring and adding Targets.&#x20;

## **Test For Specific Compliance Policy**

An option is available to [trigger a new Test Run directly from the Compliance Policy page](https://docs.probe.splx.ai/ai-red-teaming/probe/compliance#test-your-target-against-a-specific-compliance-by-initiating-a-tailored-test-run), with pre-selected Probes mapped to the specific Compliance Policy.

## Multi-step Attacks Toggle  <a href="#add-single-multi-step-attacks-toggle-button-on-target-level" id="add-single-multi-step-attacks-toggle-button-on-target-level"></a>

Users can now toggle the execution of multi-step attacks, enabling support for both conversational targets and those that only accept single-message instructions.

## New Model Benchmarks

Our [Model Benchmarks](https://docs.probe.splx.ai/ai-benchmarks/model-benchmarks) have been extended to include results for the following models:

* **Anthropic Claude 3.5 Sonnet**
* **Anthropic Claude 3 Haiku**
* **Anthropic Claude Opus 4.5**
* **cohere/command-a**
* **deepseek-ai/DeepSeek-R1-Distill-Llama-70B**
* **Gemini 3 Pro**
* **OpenAI GPT 5.2**
* **xai-org/grok-4.1-fast-reasoning**

## Improvements & Tweaks

* Transition to Zscaler branding.
* Users can now select and assign coverage levels to all probes.