September 2025

AI Asset Management

We’ve launched AI Asset Management, the discovery and management layer for all AI components inside your enterprise. It extends the SPLX Platform by connecting to environments and running scans to automatically surface models, workflows, guardrails, MCP servers, and tools across your infrastructure. With this visibility, enterprises can detect vulnerabilities, spot misconfigurations, and flag models or components as unwanted, turning the inventory into an actionable security and governance layer.

Learn more about AI Assets.

XSS Probe

• XSS Probe tests whether an AI model can be tricked into outputting script payloads like <script>alert(1)</script>, which could enable real-world cross-site scripting (XSS) attacks if rendered in a browser. The tool simulates realistic user interactions, flags identical or near-identical reproductions of malicious payloads as failures, and provides prioritized findings.

AI Runtime Protection

New guards are now available to strengthen runtime protection:

• PII Leakage Guard - Identifies and redacts sensitive personal information (like SSNs, credit cards, or phone numbers) in both user and assistant messages.

• Unverified Links Guard - Detects links that don’t match your organization’s approved domain list, helping prevent phishing and unsafe redirects.

• Custom Regex Guard - Lets you define your own rules using regular expressions to detect specific patterns, like competitor mentions or pricing details.

• Harmful Content Guard - Detects harmful or unsafe content across multiple categories (e.g. violence, self-harm, illegal actions, or sexually explicit content).

Model Benchmarks

• Performance improvements: Benchmarks now load ~10x faster.

• Export: Benchmark results can be exported as PDF for easier sharing. Learn more in the dedicated documentation section.

• Our Model Benchmarks have been extended to include results for the following models:

  • OpenAI GPT 4.1

  • OpenAI GPT 4.1-mini

  • Gemma 3 27B IT

  • Mistral Nemo

  • Apertus 70B

Test Reports

• You can now mark individual attacks to appear in generated reports (both Target & Test Run reports). Reports now include an Appendix section showing details of test cases from executed probes that were marked for inclusion.

• Error test cases are now included in generated reports for better traceability.

• Overview Page & Target Report: Data alignment improved with additional rules to prevent differences.

Public API

We’ve added a new endpoint to trigger and get results of AI Analysis.

Improvements & Tweaks

• All table rows are clickable if they have details (no more arrow hunting).

• Dates are now standardized to ISO format.